Event prediction and alarm generation method based on multi-modal physiological time sequence data

Abstract

The invention relates to the technical field of electric digital data processing, in particular to an event prediction and alarm generation method based on multi-modal physiological time sequence data, which comprises the following steps: judging whether a suspected falling trend occurs or not; determining a plurality of preliminary abnormal segments; determining a physiological feature vector corresponding to each preliminary abnormal segment; calculating a physiological comprehensive index, and determining a plurality of final abnormal segments; calculating a physiological tumble index and an environmental slip index; determining an event type corresponding to the suspected tumble trend based on the difference between the physiological tumble index and the environmental slip index; generating a corresponding alarm prompt based on the real prediction type; and adjusting a preset impact amplitude threshold value. According to the method, the problems of low prediction accuracy, high false alarm rate and poor system adaptive capacity caused by single data mode and excessive dependence on static parameters in the prior art are effectively solved through multi-mode time sequence data fusion and a dynamic threshold adjustment mechanism.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing technology, and in particular to a method for event prediction and alarm generation based on multimodal physiological time-series data. Background Technology

[0002] With the increasing aging of the global population, falls have become one of the major health risks faced by the elderly and those with mobility impairments. Falls can not only lead to physical injuries such as fractures and joint dislocations, but also increase the medical burden and affect quality of life. According to statistics from the World Health Organization, falls are one of the leading causes of accidental injuries among the elderly, thus the development of effective fall prevention technologies is an urgent social need. Traditional fall prevention measures mainly rely on manual supervision or simple sensors, but these methods suffer from problems such as insufficient real-time performance, high false alarm rates, and low personalization. In recent years, with the development of wearable devices and artificial intelligence technologies, intelligent prediction systems based on physiological data have gradually become a research hotspot. These systems collect users' physiological signals, such as heart rate and blood oxygen saturation, and environmental data, and combine them with machine learning models to achieve early prediction and alarm of fall risk, thereby improving the initiative and accuracy of prevention. However, existing technologies still have limitations in data fusion, model adaptability, and alarm mechanisms, and further optimization is urgently needed to improve reliability.

[0003] Patent document CN119970009A discloses an intelligent fall prevention prediction alarm system, comprising: a data acquisition and recording module for collecting and recording the user's first vital signs dataset and the user's real-time environmental dataset; a fall prediction module for constructing a prediction model and inputting the recorded first vital signs dataset and real-time environmental dataset into the prediction model for analysis, outputting the user's fall probability; a first alarm module for issuing a first alarm reminder when the user's fall probability reaches a preset condition; and a second alarm module for issuing a second alarm reminder after detecting a user's fall, simultaneously collecting the user's second vital signs dataset after the fall, performing secondary analysis on the second vital signs dataset based on the prediction model, and updating the user's fall probability in real time based on the analysis results.

[0004] Therefore, the existing technology has the following problems: First, the data modality is relatively simple, mainly relying on motion data such as acceleration and angular velocity and basic data such as age and fall history, but it does not integrate key physiological indicators and environmental factors, resulting in the model's inability to fully capture the multidimensional dynamic characteristics of fall risk; the model's adaptability is insufficient, relying on fixed thresholds for judgment, and lacking a dynamic adjustment mechanism based on historical data or environmental changes. In long-term use, it is prone to increased false alarm and false negative rates due to data drift or frequent interference events; the event classification is crude, only outputting alarms through a single probability value, failing to distinguish between different types such as falls due to personal imbalance and slips due to environmental factors, and lacking multi-index difference analysis to accurately identify the causes of risk, making intervention strategies untargeted and affecting nursing efficiency. Summary of the Invention

[0005] To address this, the present invention provides an event prediction and alarm generation method based on multimodal physiological time-series data. This method overcomes the problems of low prediction accuracy, high false alarm rate, and poor system adaptability in the prior art due to the single data modality, neglect of individual differences, and over-reliance on static parameters. It utilizes multimodal time-series data fusion, real-time monitoring technology, and dynamic threshold adjustment mechanism.

[0006] To achieve the above objectives, this invention provides a method for event prediction and alarm generation based on multimodal physiological time-series data, comprising: Real-time acquisition of body surface acceleration during daily activities of target individuals in elderly care facilities, and determination of whether a suspected fall trend occurs based on body surface acceleration and preset impact amplitude threshold; Based on the determination of the suspected fall trend, the angular velocity and gait symmetry of the target individual are obtained in real time, and several preliminary abnormal segments are determined according to the surface acceleration, angular velocity and gait symmetry within the preset prediction period. The heart rate variability, respiratory rate and blood pressure of the target individual within each of the preliminary abnormal segments are acquired in real time, and the physiological feature vectors corresponding to each preliminary abnormal segment are determined based on the mutation amplitude of blood pressure, heart rate variability and respiratory rate, and the synchronous mutation characteristics of heart rate variability and respiratory rate. A comprehensive physiological index is calculated based on each of the aforementioned physiological feature vectors, and several final abnormal segments are determined based on the threshold comparison results of the comprehensive physiological index. The ground friction coefficient and light intensity in the activity environment of the target individual in each of the final abnormal segments are obtained, and the physiological fall index and environmental slip index are calculated by combining the body surface acceleration and the posture angular velocity, respectively. Based on the difference between the physiological fall index and the environmental slip index, the event type corresponding to the suspected fall trend is determined to be either a true prediction type or a interference prediction type. Generate corresponding alarm prompts based on the actual prediction type; The preset impact amplitude threshold is adjusted based on the temporal distribution characteristics of the interference prediction type and the gait symmetry within the next preset observation period.

[0007] Furthermore, the process of determining whether a suspected fall tendency has occurred based on body surface acceleration and a preset impact amplitude threshold includes: When the surface acceleration exceeds the preset impact amplitude threshold, the suspected fall trend is determined to have occurred.

[0008] Furthermore, the process of determining several preliminary abnormal segments based on the body surface acceleration, posture angular velocity, and gait symmetry within a preset prediction period includes: Calculate the standard deviation of the attitude angular velocity and normalize it to obtain the normalized attitude angular velocity; The coefficient of variation of the gait symmetry is calculated and normalized to obtain the normalized gait symmetry. The fluctuation difference of the root mean square of the body surface acceleration is calculated and normalized to obtain the normalized body surface acceleration. The balance index is obtained by weighting and fusing the preset balance weights, the normalized attitude angular velocity, the normalized gait symmetry, and the normalized body surface acceleration. The preset prediction time period where the balance index is greater than the preset balance threshold is selected to determine the preliminary abnormal segment.

[0009] Furthermore, the process of determining the physiological feature vectors corresponding to each preliminary abnormal segment based on the mutation amplitude of blood pressure, heart rate variability, and respiratory rate, and the synchronous mutation characteristics of heart rate variability and respiratory rate, includes: The degree of synchronization mutation is determined based on the correlation coefficient between the heart rate variability sequence and the respiratory rate sequence within a preset sliding window. Cooperative volatility is determined based on the covariance of the heart rate variability sequence and the respiratory rate sequence; The blood pressure fluctuation range is determined based on the maximum blood pressure value and the preset blood pressure baseline; The amplitude of heart rate variability mutations was determined based on the standard deviation of the heart rate variability sequence. The magnitude of respiratory rate fluctuation is determined based on the difference between the mean respiratory rate and the preset respiratory rate baseline. Based on the synchronous mutation degree, the coordinated fluctuation degree, the blood pressure fluctuation amplitude, the heart rate variability mutation amplitude, and the respiratory rate mutation amplitude, the physiological feature vector corresponding to each preliminary abnormal segment is determined.

[0010] Furthermore, the process of determining several final abnormal segments based on the threshold comparison results of the physiological comprehensive index includes: In each of the preliminary abnormal segments, the preliminary abnormal segments whose physiological comprehensive index is greater than a preset physiological comprehensive index threshold are marked as final abnormal segments, so as to determine a number of final abnormal segments.

[0011] Furthermore, the process of obtaining the ground friction coefficient and light intensity in the activity environment of the target individual in each of the final abnormal segments, and calculating the physiological fall index and environmental slip index in combination with the body surface acceleration and the posture angular velocity, includes: The extreme tilt angle and the duration of imbalance are determined based on the final abnormal segment and the attitude angular velocity. The gait symmetry variation coefficient is determined based on the final abnormal segment and the gait symmetry. The peak impact value and waveform symmetry are determined based on the final abnormal segment and the surface acceleration. The physiological fall index is determined based on the extreme tilt angle, the duration of imbalance, the peak impact value, the waveform symmetry, and the coefficient of variation of gait symmetry. The environmental slip index is determined based on the ground friction coefficient and the impact peak value.

[0012] Furthermore, the process of determining whether the event type corresponding to the suspected fall trend is a true prediction type or a disruptive prediction type based on the difference between the physiological fall index and the environmental slip index includes: Calculate the mean, variance, maximum value, and frequency of occurrence of the physiological fall index within the preset observation period to obtain the mean, variance, maximum value, and frequency of occurrence of the physiological fall index. Calculate the mean, variance, maximum value, and frequency of occurrence of the environmental slippage index within the preset observation period to obtain the mean, variance, maximum value, and frequency of occurrence of the environmental slippage index. Calculate the correlation coefficient between the physiological fall index and the environmental slip index within the preset observation period to obtain the event correlation coefficient; The degree of difference is determined based on the mean of the physiological fall index, the mean of the environmental slip index, the variance of the physiological fall index, the variance of the environmental slip index, and the event correlation coefficient; The event type is determined based on the preset observation period, the physiological comprehensive index, the balance index, the gait symmetry, the ground friction coefficient, the impact peak value, and the light intensity.

[0013] Furthermore, the process of determining the event type based on the preset observation period, the physiological comprehensive index, the balance index, the gait symmetry, the ground friction coefficient, the impact peak value, and the light intensity includes: Based on the predicted observation period, the maximum values ​​of the physiological comprehensive index, the balance index, and the peak impact are determined respectively to obtain the maximum values ​​of the physiological comprehensive index, the balance index, and the peak impact. The maximum value of the sudden change in light intensity is determined based on the predicted observation period and the light intensity. The mean value of the ground friction coefficient is determined based on the predicted observation period and the ground friction coefficient. The maximum coefficient of variation of gait symmetry is determined based on the predicted observation period and the gait symmetry. When the difference is greater than a preset high difference threshold and the maximum coefficient of variation of gait symmetry is greater than a preset gait symmetry threshold, the event type is determined to be a true prediction type. If the difference is less than a preset low difference threshold, the average ground friction coefficient is less than a preset friction threshold, the maximum value of the impact peak is greater than a preset impact threshold, and the maximum value of the sudden change in light intensity is less than a preset light intensity threshold, the event type is determined to be an environmental slip.

[0014] Furthermore, the process of adjusting the preset impact amplitude threshold based on the temporal distribution characteristics of the interference prediction type and the gait symmetry within the next preset observation period includes: The total number of times the interference prediction type occurs within a preset unit time window is counted to determine the total number of interference prediction events per unit time. The instantaneous density is determined based on the preset unit time window and the total number of interference prediction events per unit time. The average density is determined based on the instantaneous density, the preset unit time window, and the preset observation period; The standard deviation of distribution fluctuation is determined based on the instantaneous density, the average density, the preset unit time window, and the preset observation period; The preset impact amplitude threshold is adjusted based on the average density and the standard deviation of the distribution fluctuation.

[0015] Furthermore, the process of adjusting the preset impact amplitude threshold based on the average density and the standard deviation of the distribution fluctuation includes: When the average density is greater than the preset upper limit threshold, the preset impact amplitude threshold is increased. When the average density is less than the preset lower density threshold, the preset impact amplitude threshold is lowered.

[0016] Compared with existing technologies, the beneficial effects of this invention lie in its comprehensive coverage of behavioral, physiological, and environmental features through multimodal time-series data fusion and time-series analysis. Weighted fusion based on physiological feature vectors accurately identifies real fall precursors, avoiding environmental interference. Experimental data shows that this invention improves prediction accuracy and reduces false alarm rates in elderly care scenarios. Based on the temporal distribution characteristics of interference prediction types within the next observation period, it optimizes preset impact amplitude thresholds in real time, and closed-loop feedback enables the system to adapt to individual aging or environmental fluctuations, improving long-term reliability. The calculation of a physiological fall index quantifies self-imbalance, and an environmental slip index quantifies environmental risk. Based on the difference between the two indices, it accurately distinguishes event types, and a tiered alarm mechanism improves care efficiency, avoids unnecessary emergency responses, and achieves a transformation from passive response to proactive prediction and precise intervention. This significantly improves the safety, efficiency, and economy of elderly care scenarios, effectively solving the problems of low prediction accuracy, high false alarm rates, and poor system adaptability in existing technologies due to single data modalities, neglect of individual differences, and over-reliance on static parameters. Attached Figure Description

[0017] Figure 1 This is a flowchart of the event prediction and alarm generation method based on multimodal physiological time series data in this embodiment; Figure 2 This is a flowchart for determining the physiological feature vector in this embodiment; Figure 3 This is a flowchart illustrating the calculation of the physiological fall index and the environmental slip index in this embodiment; Figure 4 This is a flowchart for adjusting the preset impact amplitude threshold in this embodiment. Detailed Implementation

[0018] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0019] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0020] Please see Figure 1 As shown, it is a flowchart of the event prediction and alarm generation method based on multimodal physiological time series data in this embodiment.

[0021] This embodiment provides a method for event prediction and alarm generation based on multimodal physiological time-series data, including: Real-time acquisition of body surface acceleration during daily activities of target individuals in elderly care facilities, and determination of whether a suspected fall trend occurs based on body surface acceleration and preset impact amplitude threshold; Based on the determination of the suspected fall trend, the angular velocity and gait symmetry of the target individual are obtained in real time, and several preliminary abnormal segments are determined according to the surface acceleration, angular velocity and gait symmetry within the preset prediction period. The heart rate variability, respiratory rate and blood pressure of the target individual within each of the preliminary abnormal segments are acquired in real time, and the physiological feature vectors corresponding to each preliminary abnormal segment are determined based on the mutation amplitude of blood pressure, heart rate variability and respiratory rate, and the synchronous mutation characteristics of heart rate variability and respiratory rate. A comprehensive physiological index is calculated based on each of the aforementioned physiological feature vectors, and several final abnormal segments are determined based on the threshold comparison results of the comprehensive physiological index. The ground friction coefficient and light intensity in the activity environment of the target individual in each of the final abnormal segments are obtained, and the physiological fall index and environmental slip index are calculated by combining the body surface acceleration and the posture angular velocity, respectively. Based on the difference between the physiological fall index and the environmental slip index, the event type corresponding to the suspected fall trend is determined to be either a true prediction type or a interference prediction type. Generate corresponding alarm prompts based on the actual prediction type; The preset impact amplitude threshold is adjusted based on the temporal distribution characteristics of the interference prediction type and the gait symmetry within the next preset observation period.

[0022] It is understood that the target individuals in this embodiment are elderly individuals in the elderly care facility whose safety and health status need to be monitored. These elderly individuals have problems such as gait disorder (e.g., Parkinson's disease), impaired balance, and unstable cardiovascular function (e.g., arrhythmia). Their daily activities include walking, getting up, turning around, and sitting down. During these activities, they are prone to falling due to loss of posture control or environmental disturbances such as slippery ground.

[0023] The process of acquiring real-time body surface acceleration, posture angular velocity, and gait symmetry during the daily activities of a target individual includes: using a triaxial accelerometer such as the ADXL345 or MPU6050, with a range of ±16g and a resolution of 13 bits, integrated into wearable devices such as smart waist belts, back patches, or wristbands, preferably fixed near the center of gravity of the torso, such as the third lumbar vertebra, and worn through an elastic waistband to reduce interference from local limb movements. The sampling frequency is set to 50Hz, covering the human movement frequency from 0 to 20Hz, and the data is transmitted to the edge computing node via Bluetooth 5.0 or ZigBee. After acquiring the triaxial acceleration components from the raw data, a low-pass filter is used to filter out high-frequency noise, and a high-pass filter is used to separate the gravitational acceleration component, retaining the dynamic motion acceleration. Then, a body coordinate system is established with the vertical direction as the z-axis when standing, the horizontal forward direction as the x-axis, and the left and right directions as the y-axis. The surface acceleration is represented as a real-time modulus, which is the modulus of the three-axis dynamic components after gravity removal. A three-axis gyroscope, such as the L3G4200D, with a range of ±2000 degrees per second and a resolution of 16 bits, is integrated with the accelerometer in the same inertial measurement unit (IMU). Data is collected through wearable devices or environmental nodes connected to the surface accelerometer. The IMU and the accelerometer are rigidly connected to ensure that the data timestamp synchronization error is less than 10ms. The raw data acquires the three-axis angular velocity components, corresponding to the rotational speeds around the roll, pitch, and yaw axes, respectively. The attitude angular velocity is represented as the modulus of the vector sum of the three-axis angular velocities. The system primarily uses trunk motion signals collected by the surface accelerometer, combined with foot-distributed pressure sensors, such as the FSR sensor built into the insole, to assist in step event detection. The core principle is to quantify gait symmetry through stride length and left-right symmetry, requiring prior identification of the gait cycle. Gait event detection uses a 0.5 to 3 Hz bandpass filter on the vertical component acceleration after gravity removal, and an adaptive thresholding method to identify heel strike and toe-off events, marking the start and end points of the step. Step length is estimated through trunk displacement or directly measured by foot pressure sensors to determine the distance between the left and right foot strike points. Step duration is the time interval between two consecutive heel strikes of the same foot. Gait symmetry is quantified as a symmetry sequence; the arithmetic mean of the step length ratio and step duration ratio is calculated for each gait cycle as the symmetry value for that cycle, with perfect symmetry defined as 1. The coefficient of variation, the ratio of the standard deviation to the mean of all symmetry values ​​within a preset time period (e.g., 10 seconds), reflects the degree of gait disorder.

[0024] The process of acquiring heart rate variability, respiratory rate, and blood pressure for a target individual includes: Heart rate variability is acquired using a chest patch ECG sensor or a wrist photoplethysmography (PPG) sensor at a sampling frequency of 250Hz or higher. After acquiring continuous ECG signals, the interval sequence of RR (respiratory rate) – the interval of a normal heartbeat – is extracted, and its standard deviation is calculated as the heart rate variability index, reflecting autonomic nervous system regulation. Respiratory rate is acquired by detecting chest wall fluctuations or respiratory modulation components in ECG / PPG signals, such as respiratory oscillations at PPG peaks and troughs, using a chest band respiratory sensor at a sampling frequency of 1Hz. After low-pass filtering and noise reduction, the respiratory rate per minute is calculated using peak counting or sine fitting. Blood pressure is measured using an arm-type automatic cuff blood pressure monitor with timed inflation or estimated using a finger clip continuous blood pressure monitor based on pulse wave conduction time. Systolic and diastolic blood pressure are acquired, and the measurement time is recorded synchronously. All data is transmitted via Bluetooth to an edge computing node, aligned with the initial abnormal segment time period, ensuring spatiotemporal consistency of data during physiological feature vector calculation. Specifically, heart rate variability and respiratory rate are taken as the mean within the segment, and blood pressure is taken as the maximum within the segment.

[0025] The process of obtaining the ground friction coefficient and light intensity in the activity environment of a target individual includes: Ground friction coefficient acquisition methods include deploying a portable friction coefficient meter, such as a pendulum friction coefficient meter, in the activity area to directly measure the dynamic friction coefficient in contact with the ground. Light intensity acquisition uses an illuminance sensor, such as the BH1750, with a range of 1-65535 lux, installed on the ceiling or wall of the activity area to avoid shadows, and real-time light intensity values ​​are collected in lux, recording changes over different time periods. Based on the acquired data, the ground friction coefficient is taken as the measured value, and the light intensity is taken as the real-time reading from the sensor; both are aligned with the final abnormal segments.

[0026] The preset prediction period is a fixed time window used to collect and analyze the behavioral and movement data of the target individual. It serves as the basic unit for identifying initial abnormal segments, covering the complete process of a single possible postural imbalance or gait disorder, ensuring the completeness of abnormal feature extraction. It depends on the physiological characteristics of the human posture-gait cycle and the temporal distribution of abnormal events: during normal walking, one step in each direction of the gait cycle takes approximately 1-2 seconds, but the imbalance process before a fall, such as tilting, gait disorder, and impact precursors, typically lasts 5-15 seconds. Considering the slow movement characteristics of elderly people in nursing homes, it is necessary to cover the longest possible incubation period for abnormalities. According to historical case statistics, over 90% of suspected fall-related abnormal events can be captured within 10 seconds, such as the onset of imbalance, gait disorder, and impact precursors. In this embodiment, the preset prediction period is set to 10 seconds. With a sampling frequency of 50Hz, 10 seconds contains 500 sampling points, which meets the calculation requirements of features such as the standard deviation of posture angular velocity and the coefficient of variation of gait symmetry. This ensures coverage of the entire chain from "posture imbalance onset - gait disorder aggravation - impact precursors", avoids short-term missed detections, and meets the real-time early warning requirements of elderly care scenarios.

[0027] The physiological comprehensive index is calculated based on each physiological feature vector, which is the arithmetic square root of the sum of the squares of each component of the physiological feature vector. The magnitude of the physiological comprehensive index positively reflects the severity of physiological abnormalities within the initial abnormal segment. The higher the value, the higher the degree of integration of autonomic-respiratory synchronization disorder, circulatory instability, and abrupt changes in autonomic and respiratory regulation, and the stronger the correlation with fall risk.

[0028] The preset observation period is a fixed time window used for statistical analysis of the difference between the physiological fall index and the environmental slippage index, covering the complete cycle of multiple suspected fall trends for the target individual. In elderly care scenarios, the frequency of suspected fall trends among the elderly is low, averaging about 0.5-1 times per day. A 7-day period can collect at least 3-7 preliminary abnormal fragments, meeting the minimum sample size requirements for calculating statistics such as mean, variance, and correlation coefficient. The balance ability of the elderly and environmental factors such as ground humidity and light intensity fluctuate less in the short term. A 7-day period can capture gradual trends such as worsening gait disorder and increased frequency of slippery environments without compromising data timeliness due to an excessively long period. In this embodiment, the preset observation period is set to 7 days, and the accumulated multi-fragment data of 3-10 pieces can support the effective calculation of statistics, avoiding misjudgment of event type due to a single abnormal fragment and improving the accuracy of distinguishing between real and interference types.

[0029] Suspected fall trends are initially screened by setting a preset impact amplitude threshold. Preliminary abnormal segments are then identified by combining posture angular velocity and gait symmetry, filtering out interference from daily activities. Physiological signal mutations and synchronization features are extracted from these preliminary abnormal segments. The degree of imbalance is quantified using a comprehensive physiological index to determine the final abnormal segment, improving the accuracy of abnormal segment identification and reducing false alarms. The physiological fall index, which integrates tilt angle, imbalance duration, and impact peak value, quantifies self-imposed imbalance. The environmental slippage index, coupled with ground friction coefficient, light intensity, and impact peak value, quantifies environmental risk. The difference between the two indices is calculated to accurately separate the true prediction type from the interference prediction type, avoiding misclassification of environmental slippage as self-imposed imbalance-related falls, reducing unnecessary emergency interventions, and accurately identifying true falls. This ensures timely handling of high-risk events; by adjusting the preset impact amplitude threshold based on the temporal distribution characteristics and gait symmetry of the interference prediction type within the next preset observation period, it addresses the shortcomings of fixed thresholds in adapting to dynamic changes in risk, improves the accuracy of suspected fall trend judgment, reduces false alarm and false negative rates, and significantly enhances system adaptability; from surface acceleration monitoring to final abnormal segment judgment, the entire process is automated to ensure real-time early warning after a suspected fall, and implements graded interventions based on event type, improving the intervention efficiency of nursing staff, and promoting the transformation of elderly care monitoring from passive response to proactive prediction and precise intervention. It effectively solves the problems of low prediction accuracy, high false alarm rate, and poor system adaptability in existing technologies due to single data modalities, neglect of individual differences, and over-reliance on static parameters.

[0030] Specifically, the process of determining whether a suspected fall has occurred based on body surface acceleration and a preset impact amplitude threshold includes: When the surface acceleration exceeds the preset impact amplitude threshold, the suspected fall trend is determined to have occurred.

[0031] The preset impact amplitude threshold is a critical value used to determine whether a target individual is showing a suspected fall tendency. It depends on the statistical analysis of historical fall event data of the target individual in the elderly care facility. In this embodiment, by collecting the surface acceleration modulus of a large number of healthy elderly people's daily activities such as walking, getting up, sitting down, and actual fall events, it was found that the impact modulus of 95% of daily activities is less than 2.5 m / s², and the impact modulus of 90% of fall events is greater than 2.5 m / s². Therefore, the preset impact amplitude threshold is set to 2.5 m / s², which can effectively distinguish between normal activities such as jumping and carrying objects and suspected falls such as loss of balance, thus reducing false alarms.

[0032] By comparing real-time calculated body surface acceleration with a preset impact amplitude threshold, the computational load is minimal. Combined with a 50Hz sampling frequency and edge node processing, low-latency response can be achieved, quickly identifying suspected fall events. Risks are accurately differentiated, and the preset impact amplitude threshold is set based on historical data of the elderly population in nursing homes, avoiding accidental triggering during daily activities while ensuring a high detection rate of actual falls. This scientifically supports subsequent processes.

[0033] Specifically, the process of determining several preliminary abnormal segments based on the body surface acceleration, posture angular velocity, and gait symmetry within a preset prediction period includes: Calculate the standard deviation of the attitude angular velocity and normalize it to obtain the normalized attitude angular velocity; The coefficient of variation of the gait symmetry is calculated and normalized to obtain the normalized gait symmetry. The fluctuation difference of the root mean square of the body surface acceleration is calculated and normalized to obtain the normalized body surface acceleration. The balance index is obtained by weighting and fusing the preset balance weights, the normalized attitude angular velocity, the normalized gait symmetry, and the normalized body surface acceleration. The preset prediction time period where the balance index is greater than the preset balance threshold is selected to determine the preliminary abnormal segment.

[0034] When calculating the normalized posture angular velocity, the magnitude of the vector sum of the three-axis posture angular velocities, reflecting the intensity of whole-body swaying, is first calculated, and the standard deviation is calculated. Then, the normalized value is normalized with reference to the preset maximum posture sway threshold of 2 rad / s, i.e., the normalized value = standard deviation / 2, and if it exceeds 1, it is taken as 1. The preset maximum posture sway threshold is the dividing threshold between normal and abnormal swaying, and is the upper limit of the intensity of posture swaying in healthy elderly people during daily activities. It reflects the maximum acceptable fluctuation range of whole-body posture stability. It depends on the statistical data of the measured posture angular velocity of the elderly population in the elderly care facility: by collecting the three-axis posture angular velocities of a large number of healthy elderly people during daily activities and calculating their standard deviation, it was found that the standard deviation of the posture angular velocity of 95% of healthy elderly people is less than 2 rad / s, while in the process of imbalance before suspected falls, the standard deviation is generally greater than 2 rad / s. Therefore, the preset maximum posture sway threshold is set to 2 rad / s to ensure that the slight swaying during normal activities and the violent swaying during abnormal events can be accurately distinguished.

[0035] When calculating normalized gait symmetry, gait event detection is first used to identify heel strike / toe lift, obtaining the arithmetic mean of the stride length ratio (left stride length / right stride length) and step duration ratio (left step duration / right step duration) as gait symmetry. Then, the coefficient of variation (COP) of all symmetry values ​​within a preset prediction period is calculated. Normalization is then performed using a preset gait significant disorder threshold as a reference: normalized value = COP / preset gait significant disorder threshold. Values ​​exceeding 1 are rounded to 1. The preset gait significant disorder threshold reflects the significant boundary between the left and right stride length / step duration disorder levels. When the COP exceeds this threshold, the gait is determined to have entered a significantly disordered state. It depends on the statistical data of the gait symmetry variation coefficient of the elderly population in the elderly care park: by collecting a large number of healthy elderly people without gait disorders in their daily activities, the gait symmetry sequence was calculated and its variation coefficient was found to be <0.5 for 95% of healthy elderly people, while the variation coefficient was generally >0.5 in the process of imbalance before suspected fall, such as the step length fluctuating and the gait disorder aggravated. Therefore, in this embodiment, the preset gait disorder threshold is set to 0.5, which can accurately distinguish between normal gait and significantly disordered gait.

[0036] When calculating normalized surface acceleration, the root mean square (RMS) of the surface acceleration modulus is first calculated, and then the standard deviation of the fluctuation of the RMS difference within a preset moving window is calculated to reflect abrupt changes in intensity. Normalization is then performed with a preset maximum intensity fluctuation threshold as a reference: normalized value = standard deviation of fluctuation / preset maximum intensity fluctuation threshold; if the value exceeds 1, it is taken as 1. The preset maximum intensity fluctuation threshold represents the maximum acceptable range of surface acceleration intensity fluctuation in healthy elderly individuals during daily activities. It depends on the statistical analysis of measured data on the RMS fluctuation of surface acceleration among elderly individuals in elderly care facilities. By collecting the RMS of surface acceleration modulus from a large number of healthy elderly individuals during daily activities such as walking, standing up, and sitting down, and calculating the standard deviation of the fluctuation of the RMS difference of surface acceleration modulus within a preset moving window, it was found that the standard deviation of fluctuation in 95% of daily activities was less than 1.5 m / s², while in the process of imbalance suspected before a fall, the standard deviation of fluctuation was greater than 1.5 m / s². In this embodiment, the maximum threshold for intensity fluctuation is set to 1.5 m / s², which can accurately distinguish between the stable intensity of normal activities such as slowly getting up and the abrupt change in intensity before an abnormal event, such as an imbalance, thus improving the specificity of the initial abnormal segment screening. The preset moving window is a sliding time window used to calculate the root mean square fluctuation difference of body surface acceleration. It depends on the temporal characteristics of intensity changes in daily human activities and abnormal events: the gait cycle of healthy elderly people is about 1-2 seconds, and the smooth changes in exercise intensity, such as walking and getting up, are usually completed within 2 seconds, while the imbalance process suspected of being before a fall, such as rapid struggling or impact precursors, will cause a sudden change in intensity within 2 seconds. In this embodiment, the preset moving window is set to 2 seconds, which can accurately quantify the abnormal feature of sudden changes in exercise intensity and distinguish between normal activities and imbalance precursors.

[0037] The preset balance weights are a combination of coefficients used to weight and fuse normalized posture angular velocity, normalized gait symmetry, and normalized surface acceleration, reflecting the contribution of posture stability, gait regularity, and movement intensity fluctuations to balance ability. They depend on the statistical analysis of the influence of each characteristic on imbalance events in elderly people in nursing homes, and are therefore allocated according to their respective weights. In this embodiment, the preset balance weights are specifically set as follows: posture stability 0.4, gait regularity 0.3, and movement intensity fluctuations 0.3. This quantifies the differentiated influence of each characteristic on balance ability, avoids misjudgment caused by a single dominant characteristic, and the weight allocation based on physiological statistics aligns with the actual imbalance in the elderly, improving the discriminative power of the balance index.

[0038] The balance index is obtained by weighting and summing the preset balance weights, normalized attitude angular velocity, normalized gait symmetry, and normalized body surface acceleration.

[0039] A preset balance threshold is used to determine whether the balance index is in a significantly abnormal state, reflecting the degree of loss of control over postural stability and gait regularity in a target individual during a suspected fall. It depends on statistical data on postural stability and measured data on imbalance events in healthy elderly individuals. Data on the angular velocity, gait symmetry, and surface acceleration of 100 healthy elderly individuals with no balance impairment during daily activities (standing, turning, walking) were collected to calculate the balance index. It was found that 95% of the healthy elderly individuals had a balance index less than or equal to 0.6, meaning a balance index greater than 0.6 is a low-probability event and is considered significantly abnormal. In a nursing home, 50 cases of self-induced falls due to imbalance, such as gait disturbances in Parkinson's disease and postural instability due to stroke, were collected. It was found that in the final abnormal segment of all events, the maximum value of the balance index was greater than 0.6, and when the balance index was greater than 0.6, the probability of postural / gait instability was greater than 50%, such as increased body tilt and asymmetry in stride length and gait duration. In this embodiment, the preset balance threshold is set to 0.6, which can accurately identify behavioral and motor abnormalities and improve the accuracy of event type determination.

[0040] By integrating three core features—the standard deviation of posture angular velocity reflecting the intensity of whole-body posture sway, the coefficient of variation of gait symmetry reflecting the degree of stride length / step time disorder, and the root mean square fluctuation difference of body surface acceleration reflecting abrupt changes in exercise intensity—this approach covers the entire chain of balance ability elements, from posture stability to gait regularity to exercise intensity fluctuation, avoiding the limitations of single features and comprehensively capturing signs of imbalance. The original features are normalized using preset thresholds to eliminate dimensional differences. Based on preset balance weights, a multi-dimensional balance index is fused, reflecting both the influence of feature differences and outputting a single, comparable indicator. The preset balance threshold is set based on the 95th percentile of the balance ability of healthy elderly individuals. Preset prediction periods with balance indices greater than the preset balance threshold are selected as preliminary abnormal segments, using statistical evidence to distinguish normal activities from abnormal events, ensuring objective and repeatable judgments.

[0041] Please see Figure 2 As shown, this is a flowchart for determining the physiological feature vector in this embodiment.

[0042] Specifically, the process of determining the physiological feature vectors corresponding to each preliminary abnormal segment based on the mutation amplitude of blood pressure, heart rate variability, and respiratory rate, as well as the synchronous mutation characteristics of heart rate variability and respiratory rate, includes: The degree of synchronization mutation is determined based on the correlation coefficient between the heart rate variability sequence and the respiratory rate sequence within a preset sliding window. Cooperative volatility is determined based on the covariance of the heart rate variability sequence and the respiratory rate sequence; The blood pressure fluctuation range is determined based on the maximum blood pressure value and the preset blood pressure baseline; The amplitude of heart rate variability mutations was determined based on the standard deviation of the heart rate variability sequence. The magnitude of respiratory rate fluctuation is determined based on the difference between the mean respiratory rate and the preset respiratory rate baseline. Based on the synchronous mutation degree, the coordinated fluctuation degree, the blood pressure fluctuation amplitude, the heart rate variability mutation amplitude, and the respiratory rate mutation amplitude, the physiological feature vector corresponding to each preliminary abnormal segment is determined.

[0043] A preset sliding window is a fixed-length time window used to slide segment by segment across a physiological signal time series, such as a heart rate variability sequence or a respiratory rate sequence, to extract synchronous features such as correlation coefficients and covariance within a local time period. This reflects short-term changes and correlations in physiological parameters, thus identifying synchronous mutations. Its value depends on the periodic characteristics of the physiological signal and the locality of synchronous mutations: a normal adult respiratory cycle is approximately 2-3 seconds, and a 5-second window can cover three respiratory cycles, fully encompassing the respiratory-related autonomic nervous system regulation cycle, avoiding noise interference if the window is too small and loss of periodic information if it is too large. Simultaneously, synchronous mutations are often brief and sudden, requiring local window capture to avoid global averaging masking key features. In this embodiment, the preset sliding window is set to 5 seconds, which can accurately capture local synchronous mutations and improve detection sensitivity.

[0044] When calculating the synchronization mutation degree, the system slides segment by segment within the initial abnormal segment. For the heart rate variability sequence (i.e., the standard deviation of the RR interval) and the respiratory rate sequence (i.e., the number of breaths per minute) within each preset sliding window, the Pearson correlation coefficient is calculated. The absolute value of the correlation coefficient for each window is taken, and normalization is performed with the theoretical maximum value of the absolute value of the correlation coefficient of 1 as a reference. That is, the normalized value = the absolute value. If it exceeds 1, it is taken as 1. The synchronization mutation degree is obtained, which reflects the degree of local synchronization disorder between the autonomic nervous system and the respiratory system.

[0045] When calculating the covariance, the covariance is calculated for the heart rate variability sequence and respiratory rate sequence within the same preset sliding window. In this embodiment, the maximum observation value obtained from the target individual's resting state over the past 30 days is used as a reference for normalization. The maximum observation value is set to 10 ms / minute, i.e., the normalized value = absolute value of covariance / maximum observation value. If it exceeds 1, it is taken as 1. This yields the covariance, which reflects the intensity of the covariance between the autonomic nervous system and respiration. Because 30 days can smooth out temporary fluctuations, by taking the maximum value instead of the mean, it is possible to focus on the extreme risk boundary under normal conditions. The 30-day cycle covers most resting state scenarios such as rest and sleep, avoiding the raising of the threshold due to single-day abnormalities such as occasional arrhythmias.

[0046] When calculating the blood pressure fluctuation range, the average of the systolic and diastolic blood pressure values ​​within the initial abnormal segment is taken, and the preset blood pressure baseline is subtracted to obtain the blood pressure difference. The normalized value is calculated using the maximum reference value of 50 mmHg for significant blood pressure fluctuations in the elderly population: normalized value = blood pressure difference / 50. When the blood pressure difference > 20 mmHg, the blood pressure fluctuation range is determined according to the normalized value calculation formula; if the difference exceeds 1, it is taken as 1. When the blood pressure difference is less than or equal to 20 mmHg, the blood pressure fluctuation range is taken as 0. The preset blood pressure baseline is the average blood pressure of the target individual in a resting state over the previous 5 minutes, serving as the individual's normal benchmark.

[0047] When calculating the amplitude of heart rate variability mutation, the standard deviation of the heart rate variability sequence, i.e. the RR interval sequence, within the initial abnormal segment is calculated. The standard deviation is normalized with reference to the maximum reference value of 20 ms for the standard deviation of heart rate variability in healthy elderly people. That is, the normalized value = standard deviation / 20. If it exceeds 1, it is taken as 1, thus obtaining the amplitude of heart rate variability mutation.

[0048] When calculating the amplitude of respiratory rate mutation, the mean respiratory rate within the initial abnormal segment is taken, and the preset respiratory rate baseline is subtracted to obtain the respiratory rate difference. The maximum reference value of 10 breaths / minute for significant respiratory rate mutation in the elderly population is used as the reference for normalization, that is, normalized value = respiratory rate difference / 10. When the respiratory rate difference is greater than 5 breaths / minute, the respiratory rate mutation amplitude is determined based on the above normalized value. When the respiratory rate difference is less than or equal to 5 breaths / minute, the respiratory rate mutation amplitude is taken as 0. The preset respiratory rate baseline is the reference mean respiratory rate of the target individual in a resting state. The value is based on the physiological characteristics of human respiration and the actual needs of elderly care scenarios: In a resting state, the respiratory rate of healthy elderly people is relatively stable. The normal adult respiratory rate is 12-20 breaths / minute, while the elderly respiratory rate is slightly lower due to the decline in lung function, about 10-18 breaths / minute. The short-term mean within 5 minutes can be used as a physiological benchmark without external interference. In this embodiment, the arithmetic mean of the respiratory rate in a 5-minute resting state is taken as the preset respiratory rate baseline. The baseline is dynamically calculated based on the individual's resting state to avoid ignoring individual differences by using the group mean, thereby improving the specificity of mutation detection.

[0049] Finally, the five normalized components are combined in sequence to form a five-dimensional physiological feature vector, namely, synchronous mutation degree, coordinated fluctuation degree, blood pressure fluctuation amplitude, heart rate variability mutation amplitude, and respiratory rate mutation amplitude. The value of each component ranges from 0 to 1. The larger the value, the more significant the abnormality of the corresponding physiological feature.

[0050] This method captures local synchronous disturbances in heart rate variability and respiration by using synchronous mutation degree and coordinated fluctuation degree; blood pressure fluctuation amplitude reflects stress-induced instability of the circulatory system, and the mutation amplitude of heart rate variability and respiratory rate is quantified by the mutation of autonomic nervous regulation function and respiratory rhythm, respectively. Multi-dimensional complementarity avoids the limitations of single indicators and comprehensively captures the precursors of physiological instability-type falls; a significant abnormal threshold is set to retain only abnormal fluctuations related to fall risk, avoiding irrelevant interference and ensuring that the feature vector focuses on the real risk; a preset sliding window slides segment by segment within the initial abnormal segment to accurately capture brief sudden synchronous mutations: avoiding global averaging from masking key features; matching the physiological cycle ensures complete extraction of local features of autonomic nervous-respiratory coordinated regulation and improves the detection sensitivity of synchronous disturbances in the few seconds before imbalance; preset blood pressure baseline and preset respiratory rate baseline are dynamically calculated based on individual resting state to avoid the group mean ignoring individual differences.

[0051] Specifically, the process of determining several final abnormal segments based on threshold comparison results of the physiological comprehensive index includes: In each of the preliminary abnormal segments, the preliminary abnormal segments whose physiological comprehensive index is greater than a preset physiological comprehensive index threshold are marked as final abnormal segments, so as to determine a number of final abnormal segments.

[0052] The preset physiological comprehensive index threshold is a critical value used to determine whether a preliminary abnormal segment has escalated into a final abnormal segment. It depends on historical data statistics, collecting physiological characteristic vectors of healthy elderly individuals (those without a history of falls or serious chronic diseases) during daily activities, calculating the distribution of their physiological comprehensive index, and using the 95th percentile as the base threshold. In this embodiment, the preset physiological comprehensive index threshold is set to 1.2, which can accurately screen for significant physiological abnormalities, avoid false alarms from daily fluctuations in healthy elderly individuals, and improve the specificity of the judgment.

[0053] By using a preset physiological comprehensive index threshold as a boundary, only preliminary abnormal segments with a physiological comprehensive index greater than the preset threshold are marked as final abnormal segments, excluding normal fluctuations and capturing significant abnormalities. Final abnormal segments must simultaneously meet two conditions: preliminary abnormality at the behavioral-motor level (balance index > preset balance threshold) and significant abnormality at the physiological level (physiological comprehensive index > preset threshold). This dual-insurance screening achieves multimodal collaborative verification.

[0054] Please see Figure 3 As shown, it is a flowchart for calculating the physiological fall index and the environmental slip index in this embodiment.

[0055] Specifically, the process of obtaining the ground friction coefficient and light intensity in the activity environment of the target individual in each of the final abnormal segments, and calculating the physiological fall index and environmental slip index by combining the body surface acceleration and the posture angular velocity, includes: The extreme tilt angle and the duration of imbalance are determined based on the final abnormal segment and the attitude angular velocity. The gait symmetry variation coefficient is determined based on the final abnormal segment and the gait symmetry. The peak impact value and waveform symmetry are determined based on the final abnormal segment and the surface acceleration. The physiological fall index is determined based on the extreme tilt angle, the duration of imbalance, the peak impact value, the waveform symmetry, and the coefficient of variation of gait symmetry. The environmental slip index is determined based on the ground friction coefficient and the impact peak value.

[0056] The process of determining the extreme value of the tilt angle and the duration of imbalance based on the final abnormal segment and attitude angular velocity includes: integrating the magnitude of the vector sum of attitude angular velocities within a preset integration window to obtain the attitude angle, i.e., the angle between the body and the vertical direction, as the tilt angle; taking the maximum value of the tilt angle within the final abnormal segment as the extreme value of the tilt angle; and summing the continuous time when the tilt angle within the final abnormal segment is greater than the preset imbalance threshold to obtain the duration of imbalance.

[0057] The preset integration window is the time window used to integrate the magnitude of the sum of the attitude angular velocities. It is determined based on the physiological characteristics of human posture changes, the adaptability of sampling frequency, and the balance between calculation accuracy. If a suspected fall is imminent, the tilting intensifies, and the attitude angular velocity accumulates a significant angle change within the preset integration window to accurately capture this instantaneous tilting trend. This avoids smoothing the angle due to an excessively long window, which could mask key signs of imbalance. In this embodiment, the preset integration window is set to 0.1 seconds. The attitude angular velocity is collected by a gyroscope with a sampling frequency of 50Hz. The 0.1-second window corresponds precisely to 5 consecutive sampling points, effectively approximating the angular velocity integration, i.e., the attitude angle change, over a short period. Experimental verification shows an error of <0.5°. The integration calculation only requires processing a small number of data points, and edge nodes can be completed in real time with a latency of <10ms, meeting the real-time requirements of elderly care scenarios.

[0058] The preset imbalance threshold is a critical value used to determine whether the tilt angle obtained after integrating the angular velocity of a target individual's posture has entered an imbalance state. It depends on the physiological statistics of the postural stability of healthy elderly people and the angular characteristics of pre-imbalance signs. By collecting a large amount of postural angular velocity data of healthy elderly people standing, turning, and walking without balance obstacles, the tilt angle was obtained by integration. It was found that 95% of healthy elderly people had a tilt angle of less than 10° when standing. In imbalance events suspected of being falls, such as increased body tilt and loss of gait control, the tilt angle is greater than 15°. In this embodiment, the preset imbalance threshold is set to 15°, which can accurately distinguish between normal and imbalanced postures and achieve accurate identification of imbalance state.

[0059] The process of determining the impact peak and waveform symmetry based on the final abnormal segment and body surface acceleration includes: taking the maximum value of body surface acceleration within the final abnormal segment as the impact peak; taking the maximum value of body surface acceleration modulus within the final abnormal segment period as the impact time; taking the acceleration modulus sequences within a preset impact duration threshold before the impact time and within the preset impact duration threshold before the impact time respectively; reversing the subsequence within the preset impact duration threshold before the impact time of the waveform to obtain the reverse impact sequence; normalizing the waveform before the impact and the reverse impact sequence to zero mean and unit variance to ensure that the similarity calculation is not affected by the absolute amplitude; and using normalized correlation to calculate the cosine similarity of the two normalized sequences to obtain the waveform symmetry.

[0060] The preset physiological fall weights include corresponding weights for extreme tilt angle, duration of imbalance, peak impact, waveform symmetry, and coefficient of variation of gait symmetry. The weight allocation for each parameter is determined through multiple regression based on the actual impact of each parameter in the imbalance event and measured data from elderly care scenarios. In this embodiment, the extreme tilt angle has a weight of 0.3. Measured data shows that the probability of falling increases by 80% when the tilt angle is >20°, hence it is given the highest weight. The peak impact has a weight of 0.25, reflecting the maximum modulus of surface acceleration at the moment of fall. Statistical analysis shows that the risk of fracture increases significantly when the peak impact is >4 m / s², hence its second highest weight. The duration of imbalance has a weight of 0.2, which is an indicator of the duration of balance regulation failure. In actual measurements, the probability of falling reaches 70% when imbalance lasts >3 seconds, hence its third highest weight. The waveform symmetry has a weight of 0.15, reflecting the symmetry of the acceleration waveform before and after the impact, and is an indirect indicator of the stability of the impact process. The more asymmetrical the waveform, the higher the risk of falling, but the impact is weaker than the previous three. The coefficient of variation for gait symmetry, weighted at 0.1, is an auxiliary indicator of gait abnormalities prior to imbalance. Gait disorder is one of the causes of falls, but it contributes the least to the immediate risk of falls.

[0061] The physiological fall index is determined by weighting and summing the extreme values ​​of tilt angle, duration of imbalance, peak impact, waveform symmetry, coefficient of variation of gait symmetry, and preset physiological fall weights. When each parameter exceeds its corresponding threshold, it is set to 1.

[0062] The Environmental Slip Index is determined based on the ground friction coefficient and peak impact value, using the following formula: ; Where s is the environmental slippage index, m is the preset minimum reference value of friction coefficient, c is the preset maximum reference value of impact peak, and f and p are the mean ground friction coefficient and impact peak value during the final abnormal segment period, respectively.

[0063] The preset minimum reference value for the coefficient of friction is a critical value used for calculating the environmental slip index, representing the lower limit of the coefficient of friction in extremely low-friction environments prone to slipping, such as wet and slippery surfaces. Experimental data, collected in elderly care facilities under extreme conditions such as wet bathrooms and oily kitchen floors, revealed that 90% of wet and slippery surfaces had a coefficient of friction ≤0.2, and slip accidents accounted for 75% of these environments. In this embodiment, 0.2 is used as the preset minimum reference value for the coefficient of friction, achieving precise quantification of environmental slip risk.

[0064] The preset maximum reference value for the impact peak is a critical value used to normalize the impact peak, mapping the actual impact peak to the [0,1] interval to quantify the relative magnitude of the impact intensity. It depends on the statistical analysis of impact peak values ​​in fall events among the elderly and measured data from extreme scenarios: by collecting surface acceleration data from real fall events among the elderly in nursing homes, it was found that when the impact peak is >4 m / s², the risk of fracture increases significantly, indicating that impacts above 4 m / s² are high-risk impacts. In severe fall scenarios such as violent struggles and head impacts, the measured maximum impact peak can reach 5 m / s², which is the upper limit of extreme impacts. In this embodiment, 5 m / s² is used as the preset maximum reference value for the impact peak, which can both cover extreme scenarios and avoid numerical overflow.

[0065] By using multiple regression analysis based on fall events in the elderly population, a preset physiological fall weight is determined to ensure that the contribution of each parameter is consistent with the actual risk. Based on the posture statistics of healthy elderly people, a preset imbalance threshold is determined to filter out minor daily swaying and retain only significant imbalance. Parameters exceeding the threshold are truncated to 1 to avoid interference from extreme values ​​and ensure that the index focuses on common high-risk scenarios. An environmental slippage index is determined based on the friction coefficient and impact peak value, accurately mapping the dual-factor synergistic nature of slippage. When both factors work together, the risk of environmental slippage is non-linearly superimposed, rather than the linear judgment of traditional single indicators. This perfectly adapts to high-frequency fall scenarios such as low friction and high impact, avoiding the one-sided problem of overlooking impact risk by only looking at friction or ignoring environmental slipperiness by only looking at impact. A high physiological fall index value indicates that self-imbalance is dominant, requiring priority intervention in balance training and fall prevention. A high environmental slippage index value indicates strong environmental slipperiness and impact, which can accurately distinguish between self-imbalance and environmental slippage, and provide caregivers with dual evidence of risk causes and intervention strategies, ultimately reducing false alarm rates, improving early warning timeliness, and ensuring the safety of the elderly.

[0066] Specifically, the process of determining whether the event type corresponding to the suspected fall trend is a true prediction type or a disruptive prediction type based on the difference between the physiological fall index and the environmental slip index includes: Calculate the mean, variance, maximum value, and frequency of occurrence of the physiological fall index within the preset observation period to obtain the mean, variance, maximum value, and frequency of occurrence of the physiological fall index. Calculate the mean, variance, maximum value, and frequency of occurrence of the environmental slippage index within the preset observation period to obtain the mean, variance, maximum value, and frequency of occurrence of the environmental slippage index. Calculate the correlation coefficient between the physiological fall index and the environmental slip index within the preset observation period to obtain the event correlation coefficient; The degree of difference is determined based on the mean of the physiological fall index, the mean of the environmental slip index, the variance of the physiological fall index, the variance of the environmental slip index, and the event correlation coefficient; The event type is determined based on the preset observation period, the physiological comprehensive index, the balance index, the gait symmetry, the ground friction coefficient, the impact peak value, and the light intensity.

[0067] Calculate the Pearson correlation coefficient between the physiological fall index and the environmental slip index within the preset observation period to obtain the event correlation coefficient.

[0068] The degree of difference is determined based on the mean of the physiological fall index, the mean of the environmental slip index, the variance of the physiological fall index, the variance of the environmental slip index, and the event correlation coefficient. Specifically, the formula for calculating the degree of difference is as follows: ; Where m is the mean of the physiological fall index, n is the mean of the environmental slip index, σ1 is the variance of the physiological fall index, σ2 is the variance of the environmental slip index, ρ is the event correlation coefficient, w1 is the preset absolute difference weight of the mean, w2 is the preset relative difference weight of the variance, w3 is the preset non-correlation weight, and ε is the preset control difference, set to 0.01 to avoid the denominator being 0.

[0069] The pre-set weights for the absolute difference of the mean, the relative difference of the variance, and the non-correlation weights are statistically analyzed based on the contribution of risk dominance and independence to event type determination. The mean difference directly reflects which risk, physiological or environmental, is more frequent and severe within the pre-set observation period, serving as the core basis for distinguishing between the true prediction type (self-imbalance-dominated) and the interfering prediction type (environmental risk-dominated). The event correlation coefficient reflects the synchronicity of the two types of risks, 1−|ρ| quantifies independence, and is a key indicator for distinguishing between risk mixtures, such as synchronous and independent physiological and environmental risks. The variance difference reflects the difference in risk fluctuation patterns, but its impact on type determination is weaker than that of the mean difference and non-correlation; fluctuation differences do not directly determine the dominant type. These findings were verified through multiple regression analysis and measured data from elderly care scenarios. Multiple regression showed that the mean difference contributes 40% to event type determination, non-correlation contributes 40%, and variance contributes 20%. In this embodiment, the preset absolute difference weight of the mean is set to 0.4, the preset relative difference weight of the variance is set to 0.2, and the preset non-correlation weight is set to 0.4. This can balance the contribution of multi-dimensional differences and improve the accuracy of event type determination.

[0070] By employing a dual-index and differential quantification approach using the physiological fall index and environmental slippage index, this method fundamentally distinguishes between two types of triggers. The physiological fall index integrates tilt angle, duration of imbalance, and peak impact to quantify the degree of imbalance; the environmental slippage index couples friction coefficient and peak impact to quantify the intensity of environmental risk. The differential quantification comprehensively considers mean difference, variance difference, and correlation coefficient to accurately capture the independence of the two indices, improving the differentiation accuracy compared to a single indicator. The method calculates the mean, variance, maximum value, and frequency of occurrence of both indices, providing a complete description of risk distribution from four perspectives, avoiding the limitations of a single indicator. Based on the Pearson correlation coefficient, the method quantifies the linear correlation between the physiological fall index and the environmental slippage index, revealing the dynamic relationship between imbalance and environmental risk in elderly care scenarios, enhancing the differential quantification capability, and avoiding interference from subjective experience. This guides precise intervention decisions; the judgment results are directly linked to intervention strategies, achieving precise risk-intervention matching and guiding precise intervention, moving from passive alarm to proactive cause identification.

[0071] Specifically, the process of determining the event type based on the preset observation period, the comprehensive physiological index, the balance index, the gait symmetry, the ground friction coefficient, the peak impact value, and the light intensity includes: Based on the predicted observation period, the maximum values ​​of the physiological comprehensive index, the balance index, and the peak impact are determined respectively to obtain the maximum values ​​of the physiological comprehensive index, the balance index, and the peak impact. The maximum value of the sudden change in light intensity is determined based on the predicted observation period and the light intensity. The mean value of the ground friction coefficient is determined based on the predicted observation period and the ground friction coefficient. The maximum coefficient of variation of gait symmetry is determined based on the predicted observation period and the gait symmetry. When the difference is greater than a preset high difference threshold and the maximum coefficient of variation of gait symmetry is greater than a preset gait symmetry threshold, the event type is determined to be a true prediction type. If the difference is less than a preset low difference threshold, the average ground friction coefficient is less than a preset friction threshold, the maximum value of the impact peak is greater than a preset impact threshold, and the maximum value of the sudden change in light intensity is less than a preset light intensity threshold, the event type is determined to be an environmental slip.

[0072] Drastic changes in light intensity, such as sudden bursts of bright light or shadow, can easily cause temporary visual blindness, leading to misjudgments of slip risk. Therefore, when determining the typical type of slip in a given environment, such interference must be eliminated first. This requires relatively stable light intensity, meaning the maximum value of the sudden change in light intensity is less than a preset threshold, thus confirming that the slip risk is directly caused by a sustained low-light environment. In this embodiment, based on the predicted observation period and light intensity, the maximum value of the sudden change in light intensity is determined by calculating the changes in light intensity at adjacent time points and taking the maximum value of each change. The arithmetic mean of the ground friction coefficients for all final abnormal segments is calculated to obtain the mean ground friction coefficient. The gait symmetry variation coefficient is calculated for the final abnormal segment, and the maximum value of all gait symmetry variation coefficients is taken to obtain the maximum gait symmetry variation coefficient.

[0073] The preset high and low thresholds for difference are two critical values ​​used to determine the difference between the physiological fall index and the environmental slip index. Based on the range of difference values, the distinction between the two types of risks is divided into three intervals: significant difference, moderate difference, and low difference, providing a quantitative standard for event type determination. This depends on the statistical analysis of measured data in elderly care scenarios and multiple regression analysis, with the core objective of maximizing the accuracy of differentiation and minimizing the false positive rate. One hundred real fall events involving elderly people in elderly care facilities were collected, with 50 cases dominated by self-imbalance and 50 cases dominated by environmental slip. The difference for each event was calculated, and a histogram of the difference distribution was plotted. The difference for the actual predicted type, i.e., self-imbalance-dominated, was concentrated in the range [0.6, 2.0], while the difference for the environmental slip type was concentrated in the range [0, 0.4]. In this embodiment, the preset high threshold for difference is set to 0.6, and the preset low threshold for difference is set to 0.4, which can improve the accuracy of event type determination and support precise intervention decisions.

[0074] The preset gait symmetry threshold is a critical value used to determine whether the coefficient of variation of gait symmetry is in a significantly disordered state. It depends on gait stability statistics of healthy elderly individuals, measured data of imbalance events, and multiple regression analysis. Step length / step time ratio data from daily walking of 100 healthy elderly individuals without gait disorders were collected, and the coefficient of variation of gait symmetry was calculated. It was found that 95% of the healthy elderly individuals had a gait symmetry degree less than or equal to 0.5. In elderly care facilities, 50 cases of self-instability-driven falls, such as those caused by Parkinson's disease, stroke, and age-related gait disorder, were collected. It was found that in the final abnormal segments of all events, the maximum value of the coefficient of variation of gait symmetry was greater than 0.5, and a gait symmetry degree greater than 0.5 indicated a significant precursor to gait imbalance. Regression was performed with the accuracy of the true prediction type as the dependent variable and the coefficient of variation of gait symmetry as the independent variable. The results showed that when the threshold was set to 0.5, the model had the highest accuracy and the lowest misjudgment rate. In this embodiment, the preset gait symmetry threshold is set to 0.5, which can accurately identify gait disorder precursors, improve the accuracy of event type determination, and guide precise intervention decisions.

[0075] The preset friction threshold is a critical value used to determine whether the average ground friction coefficient is at a level that increases the risk of slipping. It depends on measured environmental risk data and statistical analysis of slip accidents in elderly care settings. Ground friction coefficients in extremely slippery environments were collected in elderly care facilities, revealing that 90% of slippery surfaces had a friction coefficient ≤0.2, and slip accidents in these environments accounted for 75% of all slip accidents. Through controlled experiments simulating slip scenarios with different friction coefficients, it was found that when the ground friction coefficient was less than 0.3, the probability of slipping among the elderly significantly increased to over 60%, while when the friction coefficient was greater than or equal to 0.3, the probability of slipping was less than 20%. Using the accuracy of slip type identification as the dependent variable, the effects of different thresholds (0.2, 0.3, 0.4) were tested: when the preset friction threshold = 0.3, the accuracy of slip type identification was the highest, and the misjudgment rate was the lowest. If the threshold was less than 0.3, some mildly slippery but manageable environments were misjudged as high-risk; if the threshold was greater than 0.3, moderately slippery scenarios were missed. In this embodiment, the preset friction threshold is set to 0.3, which can accurately identify the risk of slippery environment, improve the accuracy of event type determination, and guide precise environmental intervention.

[0076] The preset impact threshold is used to determine whether the maximum impact peak value is at the critical value of high-risk dynamic stimulation, reflecting whether the impact intensity of the target individual in a suspected fall trend is sufficient to cause a slip or fall injury. It depends on measured impact data of elderly people falling, the correlation between slip kinetic energy and injury risk, and the accuracy verification of event type determination. In a nursing home, 100 environmental slip-dominant events were collected, and it was found that 90% of the events had impact peak values ​​greater than 3 m / s², with 85% of the impact peak values ​​falling within the 3-5 m / s² range. Biomechanical simulation of kinetic energy transfer during elderly people slipping revealed that when the impact peak value is greater than 3 m / s², the ground reaction force is sufficient to overcome static friction, especially when the coefficient of friction is less than 0.3, leading to loss of control and slipping. Furthermore, when the impact peak value is greater than 3 m / s², the risk of soft tissue contusion is greater than 50%. Using the accuracy of environmental slip type identification as the dependent variable, the effects of different thresholds of 2 m / s², 3 m / s², and 4 m / s² were tested. The highest accuracy and lowest false positive rate were observed when the preset impact threshold was 3 m / s². In this embodiment, setting the preset impact threshold to 3 m / s² accurately identifies high-impact slip risks, improves the accuracy of event type identification, and guides precise injury prevention interventions.

[0077] A preset light intensity threshold is used to determine whether the maximum value of a sudden change in light intensity is at the critical point of a stable low-light state. This reflects whether the light intensity in the target individual's activity environment is so dim that it affects visual judgment and exacerbates the risk of slipping. It depends on the experimental measurement of the visual-slip risk correlation in elderly care scenarios, statistical analysis of light in healthy environments, and verification of the accuracy of event type determination. In elderly care parks, slip accidents account for 65% in areas with light intensity less than 200 lux, higher than the 20% in areas with light intensity greater than or equal to 200 lux. In 100 dominant environmental slip events, 80% of the events occurred in environments with light intensity less than 200 lux. Using the accuracy of environmental slip type determination as the dependent variable, the effects of different thresholds of 150 lux, 200 lux, and 250 lux were tested. When the light intensity was 200 lux, the determination accuracy was the highest and the false positive rate was the lowest. In this embodiment, the preset light intensity threshold is set to 200 lux, which can accurately identify the risk of slipping in dim environments, improve the accuracy of event type determination, and guide precise lighting intervention.

[0078] By using a differential-based preliminary screening, the system initially categorizes events into two main types: "self-imbalance-driven" and "environmentally risk-driven," avoiding redundancy in full-parameter calculations. Multi-parameter extreme / mean value co-validation improves the accuracy of event type differentiation and reduces the misjudgment rate. Maximum and mean values ​​capture risk characteristics across different time periods, avoiding the randomness of single data segments and achieving dual coverage of extreme and average risks. All thresholds are set based on measured data from elderly care scenarios and statistical regression, ensuring the judgment logic is interpretable and traceable, directly guiding intervention. Through a closed-loop logic of quantifying risk independence through differential measurement, characterizing features through multi-parameter extreme / mean values, and validating with a scientific threshold system, the system achieves accurate differentiation between real-world prediction types and environmental slip-and-fall types. This not only improves judgment accuracy and reduces misjudgment rate but also, through the direct correlation between parameters, thresholds, and intervention, promotes a shift in elderly care from passive alarm to proactive cause identification and precise intervention, providing implementable, interpretable, and optimizable core technical support for smart elderly care safety management.

[0079] Please see Figure 4 As shown, it is a flowchart of adjusting the preset impact amplitude threshold in this embodiment.

[0080] Specifically, the process of adjusting the preset impact amplitude threshold based on the temporal distribution characteristics of the interference prediction type and the gait symmetry within the next preset observation period includes: The total number of times the interference prediction type occurs within a preset unit time window is counted to determine the total number of interference prediction events per unit time. The instantaneous density is determined based on the preset unit time window and the total number of interference prediction events per unit time. The average density is determined based on the instantaneous density, the preset unit time window, and the preset observation period; The standard deviation of distribution fluctuation is determined based on the instantaneous density, the average density, the preset unit time window, and the preset observation period; The preset impact amplitude threshold is adjusted based on the average density and the standard deviation of the distribution fluctuation.

[0081] A preset unit time window is a baseline unit for dividing the next preset observation period into equal-length sub-periods. Its value depends on a balance between data granularity, computational efficiency, and statistical significance. Specifically, a window that is too short will result in an excessive number of windows, increasing data storage and computational burden; a window that is too long will mask risk fluctuations in local time periods. The length of the preset unit time window must be an integer multiple of the preset observation period to ensure that there is no remaining time in the window at the end of the period. In this embodiment, the preset unit time window is set to 1 hour, which can capture hourly-level risk concentrations while effectively controlling the number of windows, achieving a balance between risk fluctuation capture and computational efficiency.

[0082] In this example, the instantaneous density is the ratio of the total number of predicted disturbance events per unit time to the window length. This allows us to obtain the instantaneous density sequence for the next preset observation period. The average density is the arithmetic mean of the instantaneous densities of all windows in the next preset observation period. Based on the instantaneous density sequence, the standard deviation of the distribution fluctuation is calculated using the standard deviation formula.

[0083] The average density reflects the overall frequency of occurrence of the interference prediction type within the next preset observation period, and the standard deviation of the distribution fluctuation reflects the stability of the occurrence of the interference prediction type. The mean of the gait symmetry variation coefficient is introduced to reflect the overall degree of gait disorder, correcting abnormal surface acceleration caused by self-imbalance and reducing false alarms due to non-fall impacts. By adjusting the temporal distribution and gait characteristics in two dimensions, a balance between sensitivity and specificity is achieved, improving the accuracy of suspected fall trend judgment and reducing false alarm and false negative rates.

[0084] Specifically, the process of adjusting the preset impact amplitude threshold based on the average density and the standard deviation of the distribution fluctuation includes: When the average density is greater than the preset upper limit threshold, the preset impact amplitude threshold is increased. When the average density is less than the preset lower density threshold, the preset impact amplitude threshold is lowered.

[0085] The preset upper and lower density thresholds are critical values ​​used to determine the risk range of the average density. Based on the temporal distribution characteristics of the interference prediction type within the next preset observation period, they provide a basis for the dynamic adjustment of the preset impact amplitude threshold. Data from 100 interference prediction type events collected in elderly care facilities revealed that when the average density is greater than 0.15 events / hour, it is easily misjudged as a suspected fall trend, requiring an increase in the threshold filtering. Using the false alarm rate as the dependent variable, different thresholds (0.1 / 0.15 / 0.2 events / hour) were tested. The preset upper density threshold of 0.15 events / hour showed the lowest false alarm rate and the highest accuracy. When the average density is less than 0.05 events / hour, it is prone to missed detections, requiring a decrease in the threshold capture rate. Using the missed detection rate as the dependent variable, different thresholds of 0.03 / 0.05 / 0.07 events / hour were tested. The preset lower density threshold of 0.05 events / hour showed the lowest missed detection rate and the highest sensitivity. In this embodiment, the preset upper limit threshold for density is set to 0.15 times / hour, and the preset lower limit threshold for density is set to 0.05 times / hour, which can improve the accuracy of judgment and reduce the false alarm and false negative rates.

[0086] When the average density exceeds a preset upper limit threshold, the risk of slipping and falling in the environment is frequent, and the threshold needs to be increased to reduce false alarms caused by high-frequency, low-intensity impacts. The adjustment range is calculated using the following formula: ; in, The upward adjustment range is given by k1, which is a preset adjustment coefficient. ρ represents the average density, and ρ is the preset upper limit threshold for density.

[0087] The preset adjustment coefficient is a fixed constant used to quantify the proportional relationship between the degree of deviation and the threshold adjustment amount. Time-series distribution characteristics of interference prediction types and threshold adjustment effects were collected from 100 elderly people in a retirement community, and the coefficients were obtained through linear regression fitting. The effects of different coefficient values ​​k1=0.3 / 0.5 / 0.7 on the false alarm rate and false negative rate were tested, and it was found that k1=0.5 can reduce both the false alarm rate and false negative rate. In this embodiment, the preset adjustment coefficient is set to 0.5 times / hour.

[0088] When the average density is less than the preset lower density threshold, the preset impact amplitude threshold is lowered. The adjustment amplitude is calculated using the following formula: ; in, The downward adjustment range is given by k, which is a preset adjustment coefficient. ρ represents the average density, and ρ is the preset lower limit threshold for density.

[0089] It is understandable that when the average density is greater than the preset upper limit threshold or when the average density is less than the preset lower limit threshold, the new preset impact amplitude threshold can be obtained by adding the upward adjustment to the preset impact amplitude threshold.

[0090] By adjusting the preset impact amplitude threshold based on the linkage between the average density and the preset density upper and lower limit thresholds, the preset impact amplitude threshold and risk frequency are accurately matched, reducing the false alarm rate and false negative rate, improving the accuracy of judgment, and promoting the upgrade of the smart elderly care system from passive alarm to proactive cause identification and precise early warning.

[0091] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for event prediction and alarm generation based on multimodal physiological time-series data, characterized in that, include: Real-time acquisition of body surface acceleration during daily activities of target individuals in elderly care facilities, and determination of whether a suspected fall trend occurs based on body surface acceleration and preset impact amplitude threshold; Based on the determination of the suspected fall trend, the angular velocity and gait symmetry of the target individual are obtained in real time, and several preliminary abnormal segments are determined according to the surface acceleration, angular velocity and gait symmetry within the preset prediction period. The heart rate variability, respiratory rate and blood pressure of the target individual within each of the preliminary abnormal segments are acquired in real time, and the physiological feature vectors corresponding to each preliminary abnormal segment are determined based on the mutation amplitude of blood pressure, heart rate variability and respiratory rate, and the synchronous mutation characteristics of heart rate variability and respiratory rate. A comprehensive physiological index is calculated based on each of the aforementioned physiological feature vectors, and several final abnormal segments are determined based on the threshold comparison results of the comprehensive physiological index. The ground friction coefficient and light intensity in the activity environment of the target individual in each of the final abnormal segments are obtained, and the physiological fall index and environmental slip index are calculated by combining the body surface acceleration and the posture angular velocity, respectively. Based on the difference between the physiological fall index and the environmental slip index, the event type corresponding to the suspected fall trend is determined to be either a true prediction type or a interference prediction type. Generate corresponding alarm prompts based on the actual prediction type; The preset impact amplitude threshold is adjusted based on the temporal distribution characteristics of the interference prediction type and the gait symmetry within the next preset observation period.

2. The event prediction and alarm generation method based on multimodal physiological time-series data according to claim 1, characterized in that, The process of determining whether a suspected fall has occurred based on body surface acceleration and a preset impact amplitude threshold includes: When the surface acceleration exceeds the preset impact amplitude threshold, the suspected fall trend is determined to have occurred.

3. The event prediction and alarm generation method based on multimodal physiological time-series data according to claim 1, characterized in that, The process of determining several preliminary abnormal segments based on the body surface acceleration, posture angular velocity, and gait symmetry within a preset prediction time period includes: Calculate the standard deviation of the attitude angular velocity and normalize it to obtain the normalized attitude angular velocity; The coefficient of variation of the gait symmetry is calculated and normalized to obtain the normalized gait symmetry. The fluctuation difference of the root mean square of the body surface acceleration is calculated and normalized to obtain the normalized body surface acceleration. The balance index is obtained by weighting and fusing the preset balance weights, the normalized attitude angular velocity, the normalized gait symmetry, and the normalized body surface acceleration. The preset prediction time period where the balance index is greater than the preset balance threshold is selected to determine the preliminary abnormal segment.

4. The event prediction and alarm generation method based on multimodal physiological time-series data according to claim 3, characterized in that, The process of determining the physiological feature vectors corresponding to each preliminary abnormal segment based on the mutation amplitude of blood pressure, heart rate variability, and respiratory rate, and the synchronous mutation characteristics of heart rate variability and respiratory rate, includes: The degree of synchronization mutation is determined based on the correlation coefficient between the heart rate variability sequence and the respiratory rate sequence within a preset sliding window. Cooperative volatility is determined based on the covariance of the heart rate variability sequence and the respiratory rate sequence; The blood pressure fluctuation range is determined based on the maximum blood pressure value and the preset blood pressure baseline; The amplitude of heart rate variability mutations was determined based on the standard deviation of the heart rate variability sequence. The magnitude of respiratory rate fluctuation is determined based on the difference between the mean respiratory rate and the preset respiratory rate baseline. Based on the synchronous mutation degree, the coordinated fluctuation degree, the blood pressure fluctuation amplitude, the heart rate variability mutation amplitude, and the respiratory rate mutation amplitude, the physiological feature vector corresponding to each preliminary abnormal segment is determined.

5. The event prediction and alarm generation method based on multimodal physiological time-series data according to claim 4, characterized in that, The process of determining several final abnormal segments based on threshold comparison results of physiological comprehensive index includes: In each of the preliminary abnormal segments, the preliminary abnormal segments whose physiological comprehensive index is greater than a preset physiological comprehensive index threshold are marked as final abnormal segments, so as to determine a number of final abnormal segments.

6. The event prediction and alarm generation method based on multimodal physiological time-series data according to claim 5, characterized in that, The process of obtaining the ground friction coefficient and light intensity in the activity environment of the target individual in each of the final abnormal segments, and calculating the physiological fall index and environmental slip index by combining the body surface acceleration and the posture angular velocity, respectively, includes: The extreme tilt angle and the duration of imbalance are determined based on the final abnormal segment and the attitude angular velocity. The gait symmetry variation coefficient is determined based on the final abnormal segment and the gait symmetry. The peak impact value and waveform symmetry are determined based on the final abnormal segment and the surface acceleration. The physiological fall index is determined based on the extreme tilt angle, the duration of imbalance, the peak impact value, the waveform symmetry, and the coefficient of variation of gait symmetry. The environmental slip index is determined based on the ground friction coefficient and the impact peak value.

7. The event prediction and alarm generation method based on multimodal physiological time-series data according to claim 6, characterized in that, The process of determining whether the event type corresponding to the suspected fall trend is a true prediction type or a disruptive prediction type based on the difference between the physiological fall index and the environmental slip index includes: Calculate the mean, variance, maximum value, and frequency of occurrence of the physiological fall index within the preset observation period to obtain the mean, variance, maximum value, and frequency of occurrence of the physiological fall index. Calculate the mean, variance, maximum value, and frequency of occurrence of the environmental slippage index within the preset observation period to obtain the mean, variance, maximum value, and frequency of occurrence of the environmental slippage index. Calculate the correlation coefficient between the physiological fall index and the environmental slip index within the preset observation period to obtain the event correlation coefficient; The degree of difference is determined based on the mean of the physiological fall index, the mean of the environmental slip index, the variance of the physiological fall index, the variance of the environmental slip index, and the event correlation coefficient; The event type is determined based on the preset observation period, the physiological comprehensive index, the balance index, the gait symmetry, the ground friction coefficient, the impact peak value, and the light intensity.

8. The event prediction and alarm generation method based on multimodal physiological time-series data according to claim 7, characterized in that, The process of determining the event type based on the preset observation period, the comprehensive physiological index, the balance index, the gait symmetry, the ground friction coefficient, the peak impact value, and the light intensity includes: Based on the predicted observation period, the maximum values ​​of the physiological comprehensive index, the balance index, and the peak impact are determined respectively to obtain the maximum values ​​of the physiological comprehensive index, the balance index, and the peak impact. The maximum value of the sudden change in light intensity is determined based on the predicted observation period and the light intensity. The mean value of the ground friction coefficient is determined based on the predicted observation period and the ground friction coefficient. The maximum coefficient of variation of gait symmetry is determined based on the predicted observation period and the gait symmetry. When the difference is greater than a preset high difference threshold and the maximum coefficient of variation of gait symmetry is greater than a preset gait symmetry threshold, the event type is determined to be a true prediction type. If the difference is less than a preset low difference threshold, the average ground friction coefficient is less than a preset friction threshold, the maximum value of the impact peak is greater than a preset impact threshold, and the maximum value of the sudden change in light intensity is less than a preset light intensity threshold, the event type is determined to be an environmental slip.

9. The event prediction and alarm generation method based on multimodal physiological time-series data according to claim 1, characterized in that, The process of adjusting the preset impact amplitude threshold based on the temporal distribution characteristics of the interference prediction type and the gait symmetry within the next preset observation period includes: The total number of times the interference prediction type occurs within a preset unit time window is counted to determine the total number of interference prediction events per unit time. The instantaneous density is determined based on the preset unit time window and the total number of interference prediction events per unit time. The average density is determined based on the instantaneous density, the preset unit time window, and the preset observation period; The standard deviation of distribution fluctuation is determined based on the instantaneous density, the average density, the preset unit time window, and the preset observation period; The preset impact amplitude threshold is adjusted based on the average density and the standard deviation of the distribution fluctuation.

10. The event prediction and alarm generation method based on multimodal physiological time-series data according to claim 1, characterized in that, The process of adjusting the preset impact amplitude threshold based on the average density and the standard deviation of the distribution fluctuation includes: When the average density is greater than the preset upper limit threshold, the preset impact amplitude threshold is increased. When the average density is less than the preset lower density threshold, the preset impact amplitude threshold is lowered.

Images