Control optimization method and system for anti-falling waistcoat for nursing old people

Abstract

The invention relates to the technical field of anti-falling waistcoat control optimization, in particular to an anti-falling waistcoat control optimization method and system for old people nursing. The method comprises the following steps that an inertial measurement unit and a distributed myoelectricity sensor are installed at key skeleton nodes in a nursing anti-falling waistcoat, and historical data extraction of falling behaviors of the old and monitoring of muscle stress tension are achieved; tumble behavior state data and muscle stress tension fluctuation data are extracted; deducing the foot landing imbalance variability and the inclination potential energy intensity increment based on historical data, and simulating the falling posture tendency; and according to the deduction data, carrying out partition buffer expansion control on the anti-falling waistcoat, and designing automatic firmware. The anti-falling waistcoat control technology is optimized, so that the anti-falling waistcoat control technology is more perfect.

Description

Technical Field

[0001] The present invention relates to the technical field of anti-fall vest control optimization, and in particular to a control optimization method and system for an anti-fall vest for elderly care. Background Art

[0002] Previous anti-fall vests relied on simple cushioning materials, unable to effectively adapt to the diverse behavioral patterns and falls experienced by the elderly in their daily activities, resulting in limited effectiveness in practical use. They also failed to predict and prevent falls in a timely manner. With the development of smart wearable devices, sensing technology, and artificial intelligence, integrating these technologies with anti-fall vests can significantly improve their performance and safety. For example, advanced technologies such as inertial measurement units (IMUs) and electromyographic sensors can be used to capture the elderly's movement status and muscle stress response. Through big data analysis, the likelihood of falls can be inferred, allowing preventive measures to be taken in advance. However, a traditional control optimization method for elderly care anti-fall vests suffers from inaccurate analysis of the elderly's fall posture and force, resulting in large errors in the vest's cushioning expansion. Summary of the Invention

[0003] Based on this, it is necessary to provide a control optimization method and system for an anti-fall vest for elderly care to solve at least one of the above technical problems.

[0004] To achieve the above objectives, a control optimization method for an elderly care anti-fall vest is provided, the method comprising the following steps: Step S1: Obtaining a data set of the elderly person's falling behavior status and a data set of muscle stress tension fluctuation records as reported by users; performing feature selection on the data set of the falling behavior status and the data set of muscle stress tension fluctuation records to obtain a data set of falling behavior status feature selection and a data set of muscle stress tension fluctuation record feature selection; Step S2: performing foot landing imbalance variability deduction on the selected dataset of fall behavior state features to obtain foot landing imbalance variability data; performing tilt potential energy intensity increment analysis based on the foot landing imbalance variability data to obtain tilt potential energy intensity increment data; performing fall posture tendency simulation deduction based on the selected dataset of muscle stress tension fluctuation records and the tilt potential energy intensity increment data to obtain fall posture tendency deduction data; Step S3: performing partition buffer expansion control of the anti-fall vest according to the fall posture tendency deduction data to obtain partition buffer expansion control data; performing automated firmware design based on the partition buffer expansion control data to obtain partition buffer expansion control firmware.

[0005] Preferably, step S2 includes the following steps: Step S21: performing a pre-fall cadence structure analysis on the fall behavior state feature selection data set to obtain pre-fall cadence structure data; Step S22: performing foot landing imbalance variability deduction on the fall behavior state feature selection data set based on the pre-fall cadence structure data to obtain foot landing imbalance variability data; Step S23: performing tilt potential energy intensity increment analysis based on the foot landing imbalance variability data to obtain tilt potential energy intensity increment data; Step S24: performing nonlinear coupling strength analysis between joint muscle groups on the muscle stress tension fluctuation record feature selection data set to obtain nonlinear coupling strength data between joint muscle groups; Step S25: performing a fall posture tendency simulation deduction based on the nonlinear coupling strength data between the joint muscle groups and the tilt potential energy strength increment data to obtain fall posture tendency deduction data.

[0006] Preferably, step S22 includes the following steps: Step S221: performing a cadence acceleration fluctuation analysis on the cadence structure data before the fall to obtain cadence acceleration fluctuation data; Step S222: performing disordered single / double-foot support time ratio identification on the cadence structure data before the fall based on the cadence acceleration fluctuation data to obtain disordered single / double-foot support time ratio data; Step S223: performing support base impairment assessment on the unordered data of single / double-foot support time ratio to generate support base impairment data; Step S224: Deducing the angle difference of the center of gravity disordered offset trajectory based on the support base loss data and the disordered data of the single / double-foot support time ratio to obtain the angle difference of the center of gravity disordered offset trajectory; Step S225: Derivation of the pressure distribution offset of the foot contact surface based on the angle difference of the center of gravity disordered offset trajectory to obtain the pressure distribution offset data of the foot contact surface; Step S226: Deducing the foot landing imbalance variability based on the angle difference of the center of gravity disordered offset trajectory and the pressure distribution offset data of the foot landing contact surface to obtain the foot landing imbalance variability data.

[0007] Preferably, step S23 includes the following steps: Step S231: Calculating the center of gravity horizontal / vertical displacement change ratio based on the foot landing imbalance variability data to obtain the center of gravity horizontal / vertical displacement change ratio; Step S232: performing center of gravity trajectory swing speed increment analysis on the center of gravity horizontal / vertical displacement change ratio to generate center of gravity trajectory swing speed increment data; Step S233: Derivation of ankle joint inversion / valgus moment mutation based on the center of gravity horizontal / vertical displacement change ratio and the center of gravity trajectory swing speed increment data to obtain ankle joint inversion / valgus moment mutation data; Step S234: calculating the approximate mean difference of the torque increase distribution at adjacent time scales for the ankle joint inversion / valgus torque mutation data to generate the approximate mean difference of the torque increase distribution; Step S235: performing a tilt potential energy dynamic spatial superposition integral on the center of gravity trajectory swing velocity increment data and the approximate mean difference of the torque increase distribution based on the Hamiltonian integral to obtain tilt potential energy spatial superposition data; Step S236: performing tilt potential energy intensity increment analysis based on the tilt potential energy spatial superposition data, the center of gravity trajectory swing speed increment data, and the ankle joint inversion / valgus torque mutation data to obtain tilt potential energy intensity increment data.

[0008] Preferably, step S25 includes the following steps: Step S251: performing time-delay coupling characteristic analysis on the nonlinear coupling strength data between the joint muscle groups to obtain time-delay coupling characteristic data between the joint muscle groups; Step S252: performing tension-frequency-energy ratio analysis on the nonlinear coupling strength data between the joint muscle groups based on the time-delay coupling characteristic data between the joint muscle groups to obtain the nonlinear tension-frequency-energy ratio between the joint muscle groups; Step S253: performing muscle group stretching and deviation correction reaction simulation analysis on the nonlinear tension-frequency-energy ratio between each joint muscle group to obtain muscle group stretching and deviation correction reaction data; Step S254: performing muscle group-potential energy falling posture tendency coupling analysis based on the muscle group stretching correction reaction data and the tilt potential energy intensity increment data to obtain muscle group-potential energy falling posture tendency coupling data; Step S255: performing a fall posture tendency simulation deduction based on the muscle group-potential energy fall posture tendency coupling data to obtain fall posture tendency deduction data.

[0009] Preferably, step S254 includes the following steps: The muscle group stretch correction response data is decomposed into dynamic tension gradients to obtain the axial tension component and radial tension component of each joint muscle group; Perform multi-dimensional plane projection decomposition on the tilt potential energy intensity increment data to obtain multi-dimensional tilt potential energy component data; Based on the multi-dimensional tilt potential energy component data, the nonlinear gradient change analysis of the human body tilt angle and angular acceleration is performed to generate the human body tilt angle change data and tilt angular acceleration change data respectively; The muscle group-potential energy falling posture tendency coupling analysis was performed based on the human body tilt angle change data, tilt angle acceleration change data and the axial tension component and radial tension component of each joint muscle group to obtain the muscle group-potential energy falling posture tendency coupling data.

[0010] Preferably, step S3 includes the following steps: Step S31: performing convolution processing on the fall posture tendency deduction data to obtain fall posture tendency convolution data; Step S32: normalizing the tilt potential energy intensity increment data to obtain tilt potential energy intensity increment normalized data; Step S33: performing partitioned buffer expansion control on the anti-fall vest according to the convolution data of the falling posture tendency and the normalized data of the tilt potential energy intensity increment to obtain partitioned buffer expansion control data; Step S34: performing automated firmware design based on the partition buffer expansion control data to obtain partition buffer expansion control firmware.

[0011] Preferably, step S33 includes the following steps: Step S331: performing wavelet transform on the convolution data of the falling posture tendency to obtain posture tendency frequency domain feature data; Step S332: determining the vest partition expansion priority sequence based on the posture tendency frequency domain feature data to obtain the vest partition expansion priority sequence; Step S333: performing a calculation for the partitioned airbag inflation pressure gradient distribution based on the vest partition inflation priority sequence and the tilt potential energy intensity increment normalization data to obtain a partitioned airbag pressure control parameter set; Step S334: performing partitioned cushioning expansion control of the anti-fall vest based on the partitioned airbag pressure control parameter set to obtain partitioned cushioning expansion control data.

[0012] Preferably, the present invention further provides a control optimization system for an elderly care anti-fall vest, which is used to execute the above-mentioned control optimization method for an elderly care anti-fall vest. The control optimization system for an elderly care anti-fall vest comprises: The data selection module is used to obtain the data set of the elderly's falling behavior status and the data set of muscle stress tension fluctuation records reported by users; perform feature selection on the data set of the falling behavior status and the data set of muscle stress tension fluctuation records to obtain the data set of the falling behavior status feature selection and the data set of the muscle stress tension fluctuation record feature selection; The fall posture tendency simulation module is used to deduce the foot landing imbalance variability based on the data set selected from the fall behavior state characteristics to obtain foot landing imbalance variability data; perform tilt potential energy intensity increment analysis based on the foot landing imbalance variability data to obtain tilt potential energy intensity increment data; and perform fall posture tendency simulation and deduction based on the data set selected from the muscle stress tension fluctuation record characteristics and the tilt potential energy intensity increment data to obtain fall posture tendency deduction data; The partition buffer expansion control module is used to control the partition buffer expansion of the anti-fall vest according to the falling posture tendency deduction data to obtain the partition buffer expansion control data; and to perform automated firmware design based on the partition buffer expansion control data to obtain the partition buffer expansion control firmware.

[0013] The beneficial effect of the present invention is that by installing inertial measurement units and distributed electromyographic sensors at key skeletal nodes inside the nursing anti-fall vest, the behavioral state and muscle stress tension fluctuations of the elderly can be monitored in real time. The inertial measurement unit can accurately obtain the elderly's motion state data, while the electromyographic sensor can monitor the tension and fluctuation of the muscles, thereby providing a scientific basis for subsequent fall behavior analysis and early warning. By extracting historical fall behavior state data and muscle stress tension fluctuation record feature selection data sets, a more accurate personalized analysis basis can be provided for further preventive measures, thereby improving the effectiveness of the anti-fall vest. The inertial measurement unit and the distributed electromyographic sensor are always running and recording. When the elderly person falls accidentally, the behavioral state and muscle stress tension fluctuations of the elderly person will be monitored and recorded to form historical data. By analyzing the historical fall behavior state data of the elderly, the imbalance variability when the foot lands can be deduced, which provides important forward-looking data for fall prediction. Combined with the foot landing imbalance variability data, the tilt potential energy intensity increment analysis is further performed to deduce the changes in external force received by the elderly during the fall process. These analysis results, combined with a dataset of selected muscle stress tension fluctuation record features, help more comprehensively simulate falling posture tendencies, thereby predicting falls in advance and optimizing protective strategies, enabling the anti-fall vest to effectively provide protection when a fall actually occurs. Controlling the anti-fall vest's zoned cushioning expansion based on the fall posture tendency deduction data can effectively disperse the impact force before the elderly fall, thereby reducing physical damage. This intelligent zoned cushioning expansion control technology not only adjusts the cushioning force based on different falling postures, but also allows for personalized adjustments to meet the diverse physical needs of the elderly. Combined with an automated firmware design, this ensures rapid response and flexible control of the anti-fall vest, allowing each vest to automatically adjust based on real-time data, further improving its fall prevention effectiveness. By combining this data with the control system, a more efficient fall prevention effect can be achieved, ensuring better protection for the elderly. Therefore, the present invention is an optimization process for a traditional control method of an anti-fall vest for elderly care, which solves the problem that the traditional optimization method of controlling an anti-fall vest for elderly care has inaccurate analysis of the elderly's falling posture tendency and force, thereby causing large errors in the buffering expansion of the anti-fall vest. It improves the accuracy of the analysis of the elderly's falling posture tendency and force, and reduces the error in the buffering expansion of the anti-fall vest. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A schematic flow chart of the steps of a control optimization method for an anti-fall vest for elderly care; Figure 2 for Figure 1 Detailed implementation steps of step S2 in FIG. Figure 3 for Figure 1Detailed implementation steps of step S3 in FIG. DETAILED DESCRIPTION

[0015] See also Figures 1 to 3 , a control optimization method for an elderly care anti-fall vest, the method comprising the following steps: Step S1: Obtaining a data set of the elderly person's falling behavior status and a data set of muscle stress tension fluctuation records as reported by users; performing feature selection on the data set of the falling behavior status and the data set of muscle stress tension fluctuation records to obtain a data set of falling behavior status feature selection and a data set of muscle stress tension fluctuation record feature selection; Step S2: performing foot landing imbalance variability deduction on the selected dataset of fall behavior state features to obtain foot landing imbalance variability data; performing tilt potential energy intensity increment analysis based on the foot landing imbalance variability data to obtain tilt potential energy intensity increment data; performing fall posture tendency simulation deduction based on the selected dataset of muscle stress tension fluctuation records and the tilt potential energy intensity increment data to obtain fall posture tendency deduction data; Step S3: performing partition buffer expansion control of the anti-fall vest according to the fall posture tendency deduction data to obtain partition buffer expansion control data; performing automated firmware design based on the partition buffer expansion control data to obtain partition buffer expansion control firmware.

[0016] In the embodiment of the present invention, reference Figure 1 The above is a flow chart of the steps of a control optimization method for an anti-fall vest for elderly care of the present invention. In this example, the control optimization method for an anti-fall vest for elderly care of the present invention includes the following steps: Step S1: Obtaining a data set of the elderly person's falling behavior status and a data set of muscle stress tension fluctuation records as reported by users; performing feature selection on the data set of the falling behavior status and the data set of muscle stress tension fluctuation records to obtain a data set of falling behavior status feature selection and a data set of muscle stress tension fluctuation record feature selection; In an embodiment of the present invention, a data set of characteristics of the falling behavior state of the elderly and a data set of characteristics of muscle stress tension fluctuation records are obtained from user feedback; based on the data collected by users during the use of the anti-fall vest, during the protection execution process (the fixed protection program set by the protective vest when leaving the factory will also be executed), and the historical fall event data that have occurred in the daily behavior activities of the elderly (including but not limited to sensor data collected from previous fall cases, physiological reaction records, myoelectric characteristics before and after the fall, etc.), the construction and connection structure of the anti-fall vest are specifically as follows: first, the key bone node positions in the human skeletal structure that are frequently active and highly correlated with falls are selected, including but not limited to bilateral hip joints, knee joints, ankle joints, shoulders, elbows, lumbar spine and cervical spine. At the above-mentioned key bone node positions, a high-precision three-axis inertial measurement unit (IMU) is installed by structural embedding. Each IMU contains a three-axis accelerometer, a three-axis gyroscope and a three-axis magnetometer. The measurement frequency is set to 200 times per second to ensure the timeliness and integrity of the dynamic data. Distributed electromyographic (EMG) sensors utilize silver fiber electrodes combined with medical conductive gel sheets, positioned at the corresponding major muscle groups, including the quadriceps femoris, gastrocnemius, lumbar and levator scapulae. Each sensor samples 1,000 times per second, achieves a signal-to-noise ratio greater than 30dB, and outputs a voltage range of 0 to 5 volts. The inertial measurement unit and distributed EMG sensors are connected to the control center via a low-power Bluetooth 5.0 module. The control center's embedded processor receives data from each node in real time and synchronizes the timing. Real-time behavioral status monitoring relies on kinematic time series feature recognition, combined with a threshold determination method to determine gait cycle, stride length, rise and fall time, and changes in body center of mass position. Multiple fixed thresholds, such as an acceleration threshold of ±2g and an angular velocity threshold of ±300 degrees per second, are set to identify rapid falls or abnormal excursions. Muscle stress and tension fluctuation monitoring relies on frequency domain analysis of EMG signals. Short-time Fourier transforms are used to extract the frequency energy distribution within 0.2 seconds. Muscle contraction stress is determined based on energy density changes in different frequency bands. Thresholds are set between 20μV and 200μV for graded response processing. The inertial measurement unit and distributed electromyographic sensors store all-day monitoring data for a week or a period in a storage module, where the data is archived with timestamps. In the historical data extraction stage, a dataset of feature selection of fall behavior status and a dataset of feature selection of muscle stress tension fluctuation records were extracted from a week of all-day monitoring. By setting a fixed marking time period, data segments with peak impact acceleration exceeding 4g, rapid sinking of the center of gravity of the corresponding posture, and tilt angle exceeding 45 degrees were automatically classified as fall behavior status records, and their behavior data were separated and output as independent datasets through labeling.The corresponding electromyographic signals within the behavioral time period were synchronously extracted. By extracting the maximum peak voltage, average voltage, coefficient of variation, and muscle activation delay parameters, the muscle stress-tension fluctuation records were time-aligned with the fall behavior data, outputting a unified time series feature data package. This data package contained the behavioral timestamp, spatial position changes of skeletal nodes, acceleration waveforms, angular velocity curves, and myoelectric spectrum change plots. Feature extraction methods were used to select features from both datasets. The features selected for the fall behavior feature dataset included 10 basic statistical features, including acceleration change amplitude, posture angle change rate, movement duration, and spatial displacement change. The features selected for the muscle stress-tension fluctuation record feature dataset included 12 characteristic indicators, including the rate of change of electromyographic signal amplitude, duration of electromyographic activity, root mean square value of the electromyographic signal, and maximum autocorrelation coefficient. The feature selection was performed using the mutual information method for preliminary screening, and the minimum redundancy maximum correlation (mRMR) method was used to confirm the final feature set. The fall behavior feature dataset and the muscle stress-tension fluctuation record feature dataset were output.

[0017] Step S2: performing foot landing imbalance variability deduction on the selected dataset of fall behavior state features to obtain foot landing imbalance variability data; performing tilt potential energy intensity increment analysis based on the foot landing imbalance variability data to obtain tilt potential energy intensity increment data; performing fall posture tendency simulation deduction based on the selected dataset of muscle stress tension fluctuation records and the tilt potential energy intensity increment data to obtain fall posture tendency deduction data; In this embodiment of the present invention, the foot acceleration and angular velocity trajectories from the historical fall behavior data obtained in step S1 are first used to identify foot landing event points. Each landing moment is determined by combining the minimum point of the second-order derivative of the acceleration curve and the trough value of the angular velocity. After the event point is determined, a sliding mean square error calculation is performed on the landing position over a continuous 5-second period to quantify the variability of foot landing imbalance. The output result is defined as the foot landing position fluctuation index. This fluctuation index is used to construct a tilt potential energy field, and the tilt potential energy intensity is derived by multiplying the equivalent mass center of gravity offset angle and the angle between the gravity direction and the foot landing position. During the analysis of the tilt potential energy intensity increment, the angle change curve within 500 milliseconds before the fall is selected, and a first-order difference superposition model is used to obtain the tilt potential energy change rate per unit time, which is recorded as the tilt potential energy intensity increment data. Next, the time-aligned electromyographic peak change rate from the muscle stress tension fluctuation record feature selection dataset is used. This is coupled with the tilt potential energy increment through vector weighting and analyzed to form a fall posture tendency simulation map. During the atlas establishment process, the K-means clustering algorithm is used to classify the posture combinations, output the tendency probability weight under each cluster category, and obtain the final fall posture tendency deduction data.

[0018] In another embodiment, foot strike imbalance variability was deduced from historical fall behavior data. By analyzing the cadence structure within 2 seconds prior to the fall, characteristic parameters were extracted: a gait period of 1.2 ± 0.3 seconds and a cadence of 0.83 ± 0.25 steps / second. The ratio of single-foot support time to double-foot support time was calculated. The normal value was approximately 3.5:1, but before the fall, this ratio dropped to 1.8:1, indicating decreased support stability. Analysis of plantar pressure distribution determined that the foot contact area decreased by 27%, and the center of pressure position shifted 32 mm laterally and 24 mm longitudinally compared to normal walking. Based on these parameters, the foot strike imbalance variability index was calculated to be 0.68 (normal range < 0.3). Incremental analysis of the tilt potential energy intensity based on the foot strike imbalance variability data revealed that the horizontal / vertical ratio of the center of gravity displacement increased from 0.15 (normal walking) to 0.42, and the center of gravity trajectory swing speed increased from the normal value of 0.12 m / s to 0.31 m / s. Further analysis of ankle inversion / valgus torque variations revealed a sudden change in torque amplitude of 12 N·m within 0.5 seconds before the fall, exceeding three times that of normal walking. Dynamic spatial superposition integration of five consecutive data points in the time series revealed an increment of 42 J in tilt potential energy intensity, reaching the protective triggering threshold. Analysis of muscle stress tension fluctuations based on electromyographic data revealed a 57% decrease in the co-contraction index of the latissimus dorsi and rectus abdominis muscles 0.3 seconds before the fall, and an abnormal delay of 120 ms was observed in the time-delay coupling characteristics of the iliopsoas and quadriceps femoris muscles. Analysis of the tension-frequency-energy ratio of the hip, knee, and ankle joint muscles revealed that the efficiency of the muscle stretch correction response decreased to 63% of the normal value. The dynamic tension gradient of the muscle groups was decomposed into axial tension components (187±23N) and radial tension components (94±18N). Combined with the multi-dimensional planar projection decomposition of the tilt potential energy, the simulation calculated that the body's forward tilt angle in the sagittal plane changed at a rate of 7.2° / s, and the lateral tilt angle in the coronal plane changed at a rate of 5.8° / s, with angular accelerations reaching 12.5° / s² and 9.3° / s², respectively. Combining these parameters to simulate the fall posture, the predicted fall direction was 45° to the right front, with the impact point expected to be in the right hip and wrist areas.

[0019] Step S3: performing partition buffer expansion control of the anti-fall vest according to the fall posture tendency deduction data to obtain partition buffer expansion control data; performing automated firmware design based on the partition buffer expansion control data to obtain partition buffer expansion control firmware.

[0020] In an embodiment of the present invention, anti-fall vest partition buffer expansion control is performed based on the fall posture tendency deduction data. First, the fall posture tendency deduction data is convolved with a 5th-order Gabor convolution kernel, and the window size is set to 15×15 to obtain a more detailed posture tendency feature map. The tilt potential energy intensity increment data is processed using the MinMax normalization method, and the numerical value is mapped to the range of 0-1 to facilitate subsequent calculations. By performing a 6-level wavelet decomposition on the posture tendency convolution data and using the Daubechies-4 wavelet basis to extract the frequency domain features, it is obtained that the high-frequency components are concentrated in the right front area. Based on this, the vest partition expansion priority sequence is determined as: right hip (priority 1), right front chest (priority 2), right waist (priority 3), right upper limb (priority 4). Based on the priority of each region and the corresponding normalized value of the tilt potential energy intensity increment, the inflation pressure of the zoned airbags is calculated. The right hip airbag pressure is set to 125 kPa, the right front chest airbag pressure is set to 105 kPa, the right waist airbag pressure is set to 95 kPa, and the right upper limb airbag pressure is set to 85 kPa. A zoned cushioning inflation control data table is constructed, containing the inflation timing, pressure values, and inflation rate parameters for each zone. These parameters are compiled into a 256KB binary-formatted zoned cushioning inflation control firmware. The zoned cushioning inflation control firmware is designed to analyze the force trends of body parts under different fall risk conditions and provide real-time zoned response and orderly inflation control for each cushioning zone of the vest, thereby achieving dynamic protection for key areas such as the back, flanks, and hips. This not only improves the timeliness and accuracy of the protection response, but also optimizes the energy consumption and user comfort of the cushioning system. It serves as a bridge for translating front-end data analysis results into actual protective actions, providing data support for optimizing and upgrading the inherent fall protection strategy of the anti-fall vest and serving as a basis for future feature correction and system parameter tuning.

[0021] Step S2 includes the following steps: Step S21: performing a pre-fall cadence structure analysis on the fall behavior state feature selection data set to obtain pre-fall cadence structure data; Step S22: performing foot landing imbalance variability deduction on the fall behavior state feature selection data set based on the pre-fall cadence structure data to obtain foot landing imbalance variability data; Step S23: performing tilt potential energy intensity increment analysis based on the foot landing imbalance variability data to obtain tilt potential energy intensity increment data; Step S24: performing nonlinear coupling strength analysis between joint muscle groups on the muscle stress tension fluctuation record feature selection data set to obtain nonlinear coupling strength data between joint muscle groups; Step S25: performing a fall posture tendency simulation deduction based on the nonlinear coupling strength data between the joint muscle groups and the tilt potential energy strength increment data to obtain fall posture tendency deduction data.

[0022] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes: Step S21: performing a pre-fall cadence structure analysis on the fall behavior state feature selection data set to obtain pre-fall cadence structure data; In an embodiment of the present invention, when a data set is selected for analyzing the gait frequency structure before falling based on the characteristics of the falling behavior state, the time-frequency analysis method is first used to process the collected acceleration data. The specific operation method is to intercept the 30-second time window before the fall of the inertial sensor raw data with a sampling frequency of 100 Hz, decompose it into multiple intrinsic mode functions through Hilbert-Huang transform, and extract the main modal components related to gait. The extracted signal is marked with the landing moment of each step by the zero crossing detection method, and the time interval between adjacent landing moments is calculated to obtain a gait cycle data sequence. The sequence is statistically processed, including calculating the average gait cycle of 1.35 ± 0.21 seconds, the gait frequency of 0.74 ± 0.12 steps / second, the gait cycle variation coefficient of 15.6%, and the stride of 0.42 ± 0.09 meters. The time domain characteristics of gait are extracted by wavelet transform, and the wavelet coefficient matrix of the gait time series is generated. Using a Morlet wavelet with a scale of 8 as the mother wavelet, the time-frequency characteristic spectrum of gait was calculated, showing that within the 10 seconds before the fall, the gait frequency gradually decreased from a normal 0.74 steps / second to 0.58 steps / second, with the fluctuation amplitude increasing to 23.8%. The maximum Lyapunov exponent of the gait sequence calculated using a recursive quantitative analysis method was 0.072, significantly higher than the normal gait value of 0.035, indicating decreased gait stability. These parameters were aggregated to form the pre-fall gait frequency structure data, consisting of a 32-dimensional time-domain eigenvector and a 48-dimensional frequency-domain eigenvector.

[0023] Step S22: performing foot landing imbalance variability deduction on the fall behavior state feature selection data set based on the pre-fall cadence structure data to obtain foot landing imbalance variability data; In an embodiment of the present invention, a dataset is selected based on the characteristics of the fall behavior state before the fall to deduce the variability of foot landing imbalance. First, the cadence acceleration fluctuation is analyzed, and the sliding window variance method is used to calculate the local fluctuation intensity of the acceleration signal. The window size is set to 250ms and the step length is 50ms. The standard deviation of the vertical acceleration in each window is calculated, and it is found that the acceleration fluctuation amplitude increases from 0.42m / s² during normal walking to 1.17m / s². Using the improved Poincaré plot analysis method, a scatter plot is drawn with the current gait cycle as the horizontal axis and the next gait cycle as the vertical axis. The short-term variability SD1 is calculated to be 0.15s, the long-term variability SD2 is 0.28s, and the variability ratio SD1 / SD2 is 0.54, which exceeds the 0.32 threshold of normal gait. The fundamental frequency and harmonic components of the gait are extracted by Fourier analysis, and the harmonic ratio is calculated. It is found that the fundamental frequency power decreases by 31% before the fall, and the second-order harmonic increases by 48%, indicating an abnormal gait rhythm. Based on the abnormal characteristics of cadence structure, the ratio of single- to double-support time was further analyzed. The plantar pressure sensor data was used to determine the transition time between support phases. The calculated average single-support time was 0.43±0.09 seconds, and the average double-support time was 0.29±0.07 seconds. The single- to double-support time ratio was 1.48:1, significantly lower than the normal gait value of 3.2:1. The coefficient of variation of the calculated support time ratio was 26.7%, exceeding the normal value by 2.3 times, indicating a severe disruption of the support rhythm. Combining these parameters, the change in base of support area was calculated using polynomial interpolation. The effective support area decreased from a normal value of 213 cm² to 142 cm², a reduction rate of 33.3%. The weighted center of mass method was used to calculate the change in center of gravity position during walking, and the center of gravity displacement velocity and acceleration were calculated using the five-point difference method. The analysis showed that the maximum lateral center of gravity excursion reached 7.2 cm, with a velocity of 8.6 cm / s. During normal walking, lateral excursion is typically less than 3.5 cm. Using a three-dimensional projection method to calculate the angle between the center of gravity trajectory and the ideal trajectory, the average angle difference between the disordered center of gravity deviation trajectory was 12.8°, with a maximum of 23.6°. Analysis of the pressure distribution at the foot contact surface using plantar pressure distribution mapping revealed dramatic fluctuations in the position of the plantar pressure center before the fall. The pressure center path length increased by 58%, the pressure center movement speed increased by 72%, and the pressure center trajectory coverage area increased by 127%. Combining these parameters, the foot landing imbalance variability index was calculated using a fuzzy comprehensive evaluation method and was found to be 0.76 (normal range: less than 0.3), indicating severe walking instability.

[0024] Step S23: performing tilt potential energy intensity increment analysis based on the foot landing imbalance variability data to obtain tilt potential energy intensity increment data; In the embodiment of the present invention, the tilt potential energy intensity increment is analyzed based on the foot landing imbalance variability data. First, the center of gravity projection method is used to calculate the displacement components of the body's center of gravity in the horizontal and vertical planes. The horizontal displacement speed is 0.28m / s and the vertical displacement speed is 0.07m / s through discrete differential solution. The horizontal / vertical displacement change ratio is 4.0, which is much higher than the 1.2 threshold value of stable walking. The center of gravity trajectory change curve is solved by the integral operation of the three-dimensional acceleration signal, and the trajectory curvature is calculated after smoothing using the spline interpolation method, and the average curvature is 0.073 , the maximum curvature is 0.182 The center of gravity trajectory was reconstructed in phase space, with the embedding dimension set to 4 and the time delay set to 8 sampling points. The calculated center of gravity trajectory swing velocity increment was 0.17 m / s², significantly higher than the 0.05 m / s² threshold for stable walking. The biomechanical inverse method was used to calculate the ankle inversion / valgus torque changes based on the plantar pressure distribution and lower limb joint angle data. Frame-by-frame analysis revealed that the ankle inversion torque increased sharply from 2.3 N·m to 8.7 N·m in the 0.5 seconds before the fall, with an increase rate of 278% / second. The torque data was subjected to a discrete Fourier transform to extract the spectral characteristics of the torque change. The main frequency was calculated to increase from 1.2 Hz in stable walking to 2.8 Hz, indicating that the ankle torque fluctuation was intensified. The sliding window method was used to calculate the difference in torque changes within adjacent 20 ms time windows, and the approximate mean difference of the torque increase distribution was 1.86 N·m, exceeding the 0.42 N·m during normal walking. Based on the principles of Hamiltonian mechanics, the approximate mean difference between the center of gravity trajectory velocity increment and the torque rise distribution was substituted into the kinetic-potential energy coupling equation, and the dynamic spatial superposition integral of the tilt potential energy was calculated using the Gauss-Legendre numerical integration method. The integration interval was set to 1 second before the fall, with an integration step of 10 ms. The calculated total spatial superposition value of the tilt potential energy was 65.3 J. Based on the principle of energy conservation, combined with the center of gravity trajectory velocity increment data and the ankle inversion / valgus torque mutation data, the energy distribution ratio method was used to calculate the tilt potential energy intensity increment to 47.8 J / kg, exceeding the protective threshold of 40 J / kg.

[0025] Step S24: performing nonlinear coupling strength analysis between joint muscle groups on the muscle stress tension fluctuation record feature selection data set to obtain nonlinear coupling strength data between joint muscle groups; In an embodiment of the present invention, a dataset of muscle stress tension fluctuation record features was selected to analyze the nonlinear coupling strength between joint muscle groups. First, the electromyographic signal with a sampling rate of 1000 Hz was filtered using a Butterworth bandpass filter (20-450 Hz) to eliminate motion artifacts and power frequency interference. The root mean square method was used to calculate the electromyographic signal amplitude, with a window size set to 100 ms and an overlap rate of 50%. The electromyographic signals of the anterior thigh (quadriceps femoris), posterior thigh (hamstrings), calf (gastrocnemius and tibialis anterior), and trunk core muscle group (rectus abdominis and erector spinae) were extracted respectively, and the maximum voluntary contraction percentage of each muscle group was calculated. Within 0.5 seconds before the fall, it was found that the activity of the erector spinae increased to 62% of the maximum voluntary contraction, while the activity of the rectus abdominis decreased to 28% of the maximum voluntary contraction, indicating an imbalance in the synergistic ratio of the muscle groups. Cross-correlation analysis was used to calculate the temporal coupling between muscle groups, with a maximum time lag of 200ms. The average cross-correlation coefficient between the hamstrings and quadriceps was calculated to be -0.32, indicating decreased coordination between antagonistic muscles. Bicoherence spectral analysis was used to assess the frequency-domain coupling between muscle groups. The coherence value between the hip flexors and extensors in the 8-12 Hz frequency band was calculated to be 0.42, lower than the 0.73 observed in normal walking, indicating decreased coordinated control between muscle groups. The Granger causality test was used to analyze the causal relationships between muscle groups, with a lag order of 5 and an F-test significance level of 0.05. The results showed that the causal strength index of the core muscles on the lower limb muscles was 0.37, significantly lower than the 0.68 observed in normal walking. The mutual information entropy method was used to quantify the strength of nonlinear coupling between muscle groups. The results showed that the average mutual information entropy between muscle groups before the fall was 0.42 bits, lower than the 0.78 bits observed in normal walking. Using singular value decomposition to extract the principal components of the electromyographic pattern, the contribution of the first three principal components was calculated to be 82%, higher than the 65% observed in normal walking, indicating a simplified muscle coordination pattern. Combining these indicators, a weighted summation method was used to determine the nonlinear coupling strength index between joint muscle groups to be 0.39 (normal range: 0.7-0.9), indicating a significant decrease in coordination between muscle groups.

[0026] Step S25: performing a fall posture tendency simulation deduction based on the nonlinear coupling strength data between the joint muscle groups and the tilt potential energy strength increment data to obtain fall posture tendency deduction data.

[0027] In an embodiment of the present invention, a fall posture tendency simulation is performed based on the nonlinear coupling strength data and the tilt potential energy intensity increment data between the joint muscle groups. First, the nonlinear coupling strength data between the joint muscle groups are subjected to a time-delay coupling characteristic analysis between the joint muscle groups, and the sliding cross-correlation method is used to calculate the activation timing relationship between different muscle groups. The analysis window is set to 250ms and the step length is 50ms. It is calculated that the activation of the hip flexors is delayed by 87ms compared with the extensors, the activation of the knee flexors is advanced by 65ms compared with the extensors, and the activation of the ankle dorsiflexors is delayed by 121ms compared with the plantar flexors. The time-frequency coupling strength between the muscle groups in different frequency bands is evaluated by wavelet coherence analysis. Using Morlet wavelet, the scale range is set to 1-64, and the average coherence coefficient between the muscle groups in the 8-16Hz frequency band is calculated to be 0.31, which is significantly lower than the 0.65 of normal walking. Based on the time-delay coupling characteristic data, the power spectral density of each muscle group is calculated by spectral analysis, and the tension-frequency energy ratio is calculated by the frequency band energy comparison method. The energy ratios for the hip flexors in the 20-100 Hz and 100-300 Hz frequency bands were calculated to be 1.68, for the knee extensors 2.13, and for the ankle plantar flexors 1.97. These ratios were all above the 1.2 threshold for normal walking, indicating an increased proportion of high-frequency fatigue components. A muscle dynamics model was used to simulate the responses of the muscles to sudden stretch stimulation. The stretch speed was set at 30° / s and the stretch amplitude was 15°. The myoelectric response latency and peak response intensity were recorded. Results showed that the delay time of the muscle stretch reflex before the fall was 78±12 ms, significantly prolonged compared to the normal value of 42±8 ms. The peak response intensity was 37% of the maximum voluntary contraction, significantly reduced compared to the normal value of 65%. The total energy of the muscle stretch correction response was calculated by time integration of the myoelectric response curve and was 53% of the normal level, indicating a significant weakening of the muscle protective response. Combined biomechanical analysis was used to evaluate the muscle stretch correction response data and the tilt potential energy intensity increment data. First, the dynamic tension of the muscle groups was decomposed into axial tension components (along the long axis of the bone) and radial tension components (perpendicular to the long axis of the bone). Polynomial fitting was used to calculate the sum of the axial tension components of the muscles around each joint to be 163 N, the sum of the radial tension components to be 87 N, and the axis-to-diameter ratio to be 1.87. Three-dimensional spatial decomposition of the tilt potential energy intensity increment data revealed sagittal components of 32.4 J / kg, coronal components of 26.5 J / kg, and transverse components of 8.9 J / kg. Numerical differentiation was used to calculate the rate of change of the body's tilt angle, yielding a sagittal forward tilt velocity of 8.7° / s and a coronal right tilt velocity of 6.3° / s. Angular acceleration was calculated using second-order differentiation, yielding sagittal angular accelerations of 14.2° / s² and 10.5° / s².These parameters were substituted into the multibody dynamics equations and numerically solved using the Runge-Kutta method. The time step was set to 10ms and the solution interval was 1 second. The muscle group-potential energy falling posture tendency coupling vector was obtained. Based on the coupling vector, the maximum likelihood estimation method was used to predict the fall direction to be 37° to the right front, with the main force point being the right hip and the secondary force points being the right shoulder and right wrist. The impact force was predicted to be 2.8 times the body weight and the impact time to be 0.21 seconds through collision dynamics analysis. The weighted voting method was used to comprehensively evaluate the probability distribution of various falling postures, and the final fall posture tendency deduction results were: the probability of falling to the right front was 78%, the probability of falling to the right was 15%, and the probability of falling forward was 7%.

[0028] Step S22 includes the following steps: Step S221: performing a cadence acceleration fluctuation analysis on the cadence structure data before the fall to obtain cadence acceleration fluctuation data; Step S222: performing disordered single / double-foot support time ratio identification on the cadence structure data before the fall based on the cadence acceleration fluctuation data to obtain disordered single / double-foot support time ratio data; Step S223: performing support base impairment assessment on the unordered data of single / double-foot support time ratio to generate support base impairment data; Step S224: Deducing the angle difference of the center of gravity disordered offset trajectory based on the support base loss data and the disordered data of the single / double-foot support time ratio to obtain the angle difference of the center of gravity disordered offset trajectory; Step S225: Derivation of the pressure distribution offset of the foot contact surface based on the angle difference of the center of gravity disordered offset trajectory to obtain the pressure distribution offset data of the foot contact surface; Step S226: Deducing the foot landing imbalance variability based on the angle difference of the center of gravity disordered offset trajectory and the pressure distribution offset data of the foot landing contact surface to obtain the foot landing imbalance variability data.

[0029] In an embodiment of the present invention, when analyzing the cadence acceleration fluctuation of the cadence structure data before a fall, the fast Fourier transform is first used to extract the spectral characteristics of the vertical acceleration signal. The acceleration data with a sampling frequency of 100 Hz is segmented using a Hanning window, and the window length is set to 512 sampling points with an overlap rate of 50%. The power spectrum density within each window is calculated to form a time-frequency spectrum. The cadence main frequency is identified by the spectrum peak extraction method. The main frequency of normal gait is between 0.8-1.2 Hz. Analysis of data within 30 seconds before the fall shows that the cadence main frequency gradually decreases from the initial 0.96 Hz to 0.71 Hz, and the spectrum energy distribution changes from concentrated to dispersed, the spectrum peak decreases by 27%, and the spectrum width increases by 41%. The time domain variance analysis method is used to calculate the acceleration fluctuation amplitude, using a 200 ms sliding window with a step size of 50 ms. It is found that the standard deviation of the vertical acceleration increases from 0.38 m / s² during normal walking to 1.05 m / s² before the fall. The sample entropy algorithm was used to evaluate the complexity of the acceleration signal, with an embedding dimension of 2 and a similarity tolerance of 0.15 standard deviations. The calculated sample entropy value increased from 1.28 during normal walking to 1.76 before the fall, indicating a decrease in gait regularity. The acceleration signal was decomposed into different frequency bands using wavelet decomposition, and the db4 wavelet basis function was used to decompose the signal to five levels. The energy trends of each subband were analyzed. Results showed that the energy contribution of the high-frequency component (12.5-25 Hz) increased from 8% during normal walking to 17% before the fall, indicating a decrease in gait stability. A recursive quantitative analysis method was used to calculate the deterministic index of the acceleration signal, with a threshold of 15% of the mean standard deviation, a recursive delay of 10 sampling points, and an embedding dimension of 3. The determined index decreased from 82% during normal walking to 64% before the fall. Comprehensive analysis revealed a significant increase in cadence acceleration fluctuation and a significant decrease in gait stability, with a fluctuation index of 0.68 (normal range <0.35). The single / double support time ratio was randomly identified from pre-fall cadence data based on cadence acceleration fluctuations. A hidden Markov model was used to identify the different phases of the gait cycle, with the number of states set to four (initial contact phase, load response phase, mid-stance phase, and swing phase). The training dataset consisted of 50 normal gait cycles. The test data was decoded using the Viterbi algorithm to identify the gait phase corresponding to each moment. Based on the decoding results, a threshold detection method was used to determine the plantar contact state. Contact was determined when the vertical ground reaction force exceeded 5% of body weight. Single-support time (one foot in contact with the ground) and double-support time (both feet in contact with the ground) were calculated. In normal elderly walking, single-support time accounts for approximately 38% of the gait cycle, while double-support time accounts for approximately 24%, resulting in a single / double support time ratio of approximately 1.58:1.Analysis of gait data within the 10 seconds before a fall revealed that single-leg support time decreased to 32% of the gait cycle, while double-leg support time increased to 29%, resulting in a single-leg / double-leg support time ratio of 1.10:1. The coefficient of variation method was used to assess the stability of support time, calculating the ratio of the standard deviation of support time to the mean over 10 consecutive gait cycles. The coefficient of variation was 8.2% during normal walking, but increased to 17.6% before a fall. The entropy method was used to quantitatively assess the disorder of the support time series. Using the approximate entropy algorithm with parameters m = 2 and r = 0.2 × standard deviation, the approximate entropy value increased from 0.156 during normal walking to 0.284 before a fall. The dynamic time warping algorithm was used to compare the support time pattern before a fall with the standard pattern. The Euclidean distance function was used, and a window constraint was set to 25%. The average distortion distance was 0.37 (normal value < 0.15). The cyclic autocorrelation method was used to analyze the periodicity of the support time series, with a lag of 10 gait cycles. The autocorrelation coefficient decreased significantly before the fall, with the first-order autocorrelation coefficient dropping from 0.83 during normal walking to 0.62, indicating a weakening of the periodicity of the support rhythm. Combining these indicators, the disorder index for the proportion of single- to double-support time was 0.73 (normal range: <0.4).

[0030] Base of support impairment was assessed using unordered data on single- and double-support time ratios. Plantar pressure data were first collected using a plantar pressure distribution sensor array with a resolution of 5 mm × 5 mm, a sampling frequency of 50 Hz, and a pressure range of 0–1000 kPa. The convex hull algorithm was used to calculate the plantar pressure contact area. During walking in normal elderly individuals, the base of support area during the double-support phase was approximately 325 cm², while the base of support area during the single-support phase was approximately 145 cm². Analysis of plantar pressure data before a fall showed that the base of support area during the double-support phase decreased to 265 cm², representing an 18.5% impairment; the base of support area during the single-support phase decreased to 118 cm², representing an 18.6% impairment. The second-order moment method was used to calculate the geometric characteristics of the plantar pressure distribution, including the center of mass position, principal axis orientation, and moment of inertia. The results showed that the moment of inertia ratio (the ratio of the maximum principal axis moment of inertia to the minimum principal axis moment of inertia) of the plantar pressure distribution before the fall increased from 2.3 during normal walking to 3.1, indicating that the shape of the support base became narrower. The boundary curvature analysis method was used to evaluate the complexity of the plantar contact edge, and the average boundary curvature was calculated to be 0.087 during normal walking. Down to 0.072 The boundary length decreased by 12.3%, indicating more concentrated plantar contact. Dynamic changes in plantar pressure distribution were assessed using pressure center trajectory analysis, calculating the path length and velocity of the pressure center from heel to toe. The path length of the pressure center was approximately 240 mm during normal walking, but decreased to 205 mm before the fall. The average velocity of the pressure center decreased from 375 mm / s during normal walking to 325 mm / s, indicating reduced forefoot weight bearing. The effective support area was quantified using a weighted area method, defining areas with plantar pressures greater than 20 kPa as the effective support area and assigning weights based on the pressure value. The calculated effective support area index before the fall was 74% of that during normal walking. Combining these indices, a base of support impairment index of 0.64 (normal range <0.3) was derived using fuzzy inference. The angle difference of the center of gravity's disordered excursion trajectory was deduced based on the base of support impairment data and the randomized data on the proportion of single-foot / double-foot support time. First, the body's center of mass position was calculated using the segmented centroid method. The human body was divided into 16 major segments. The position and orientation of each segment was measured using an inertial measurement unit (IMU). The coordinates of the center of mass were then calculated based on anatomical data. The center of mass trajectory data was smoothed using an unscented Kalman filter. The process noise covariance matrix was set to a diagonal matrix with diagonal elements of [0.01, 0.01, 0.01] m², and the measurement noise covariance matrix was set to a diagonal matrix with diagonal elements of [0.005, 0.005, 0.005] m². The projection trajectory of the center of mass in the horizontal plane was calculated. During normal walking, the center of mass trajectory exhibited a regular sinusoidal waveform with a left-right deviation of approximately 3.2 cm. The dynamic characteristics of the center of mass trajectory were analyzed using phase space reconstruction, with an embedding dimension of 3 and a time delay of 10 sampling points. The degree of chaos in the trajectory was assessed by calculating the maximum Lyapunov exponent of the phase space trajectory. The exponent was 0.028 during normal walking but increased to 0.076 before the fall, indicating decreased trajectory stability. The directional changes in the center of mass trajectory were calculated using a piecewise linear fitting method with a 0.5-second window size, and the difference in trajectory angular orientation between adjacent windows was calculated. The average angular orientation difference between consecutive gait cycles during normal walking was 7.5°, but increased to 16.8° before the fall. Wavelet coherence analysis was used to assess the correlation between the center of mass trajectory and base of support changes. Complex Morlet wavelets were used with a scale range of 1–64, corresponding to a frequency range of 0.1–6.4 Hz. The results showed that the coherence coefficient in the 0.5–1.5 Hz frequency band decreased from 0.78 during normal walking to 0.43 before the fall, indicating a decrease in coordination between center of mass control and base of support. Angle difference cumulative analysis was used to calculate the cumulative angular difference between the actual center of mass trajectory and the ideal trajectory, defined as a straight forward path. The cumulative angular difference during normal walking was 28° / 10 steps, which increased to 65° / 10 steps before the fall. Chaotic phase diagram analysis was used to evaluate the projection characteristics of the center of gravity trajectory in the anterior-posterior and left-right planes, and the trajectory envelope area and major-minor axis ratio were calculated.The results showed that the area of the trajectory envelope increased by 53% before the fall, and the ratio of the major axis to the minor axis decreased from 2.8 to 1.9, indicating a more circular trajectory and decreased stability. Based on these parameters, multiple regression analysis determined that the angular difference in the center of gravity shift trajectory was 17.3° (normal range <8°).

[0031] The pressure distribution shift at the foot contact surface was derived from the angle difference of the center of gravity's random deviation trajectory. First, a high-resolution plantar pressure measurement system was used to record plantar pressure distribution. The sensor array consisted of 64 × 48 pressure-sensing elements, each measuring 5 mm × 5 mm, with a sensitivity of 0.5 kPa, a dynamic range of 0–1000 kPa, and a sampling frequency of 100 Hz. A regional segmentation method was used to divide the plantar surface into seven functional regions: medial heel, lateral heel, medial midfoot, lateral midfoot, medial forefoot, mid-forefoot, and lateral forefoot. The integrated pressure value for each region was calculated to characterize the load bearing. During normal walking, the heel bears approximately 30% of the total load, the midfoot bears approximately 15%, and the forefoot bears approximately 55%. Analysis of the pressure data before a fall showed that the load bearing increased to 35% in the heel, 18% in the midfoot, and 47% in the forefoot, indicating a posterior shift of the load bearing. The dynamic characteristics of pressure distribution were evaluated using center of pressure trajectory analysis, which calculated the trajectory of the center of pressure from heel to toe. During normal walking, the center of pressure trajectory was smooth, primarily moving 0.5-1.0 cm lateral to the longitudinal axis of the foot. Prior to a fall, the center of pressure trajectory became irregular, with lateral deviation increasing to 1.5-2.5 cm and a 18% decrease in trajectory length, indicating a weakening of the forefoot push-off force. The load-bearing impulse for each region was calculated using the pressure-time integration method, with a window size set to the complete gait cycle. Results showed that the load-bearing impulse on the lateral side of the plantar increased by 32% and on the medial side decreased by 23% before a fall, indicating a lateral bias in plantar pressure distribution. Principal component analysis was used to extract the main characteristic patterns of the pressure distribution. Each pressure image frame was flattened into a vector to construct a pressure data matrix. The eigenvalues and eigenvectors of the covariance matrix were calculated. The first three principal components explained 86% of the variance during normal walking, but only 72% before a fall, indicating a more complex and variable pressure distribution. The modified Hausdorff distance method was used to calculate the difference between the pre-fall pressure distribution and the normal template, with the pressure threshold set at 5% of the maximum pressure to generate a binary pressure profile. The calculated results showed that the average distance between the pre-fall pressure distribution and the normal template was 8.5 mm, with a maximum distance of 14.3 mm, significantly exceeding the normal fluctuation range (average distance <5 mm). The Fourier descriptor method was used to analyze the shape characteristics of the plantar pressure profile, extracting the top 20 low-frequency coefficients. The results showed that the energy proportion of the high-order Fourier coefficients of the pre-fall pressure profile increased, indicating a more irregular profile shape. Based on these parameters, a support vector regression model was used to determine the foot contact pressure distribution deviation index of 0.71 (normal range <0.35). Foot landing imbalance variability was deduced based on the angle difference of the center of gravity random deviation trajectory and the foot landing contact pressure distribution deviation data. First, a multi-level evidence framework was used to integrate the aforementioned indicators, dividing each indicator into five levels of evidence: cadence variability, support time characteristics, support base characteristics, center of gravity trajectory parameters, and plantar pressure distribution parameters.The indicators within each level were integrated using the Dempster-Shafer evidence synthesis rule, with basic probability assignments based on the standardized scores of the indicators. After evidence synthesis, the support for each level was determined to be 0.72 for the cadence variability level, 0.68 for the support time characteristics level, 0.64 for the support base characteristics level, 0.76 for the center of gravity trajectory parameters level, and 0.71 for the plantar pressure distribution parameters level. The weights for each level were determined using the analytic hierarchy process (AHP) and were 0.15, 0.18, 0.22, 0.25, and 0.20, respectively. A weighted evidence synthesis method was used to calculate the comprehensive imbalance score, resulting in a preliminary footstrike imbalance variability index of 0.70. Imbalance variability was further accurately assessed using a fuzzy neural network. The inputs were the five hierarchical indicators described above. The network structure consisted of five input nodes, 10 hidden nodes, and one output node. The activation function for the hidden layer was the hyperbolic tangent function, and the activation function for the output layer was the sigmoid function. The network was trained using 128 sets of labeled samples, with a learning rate of 0.05 and a momentum factor of 0.8 for 200 training iterations. The aforementioned metrics were fed into the trained network, resulting in a foot strike imbalance variability index of 0.76 (thresholds set as: <0.3 for normal, 0.3-0.5 for mild imbalance, 0.5-0.7 for moderate imbalance, and >0.7 for severe imbalance). A time series forecasting method was used to assess imbalance trends. An autoregressive integrated moving average model with an ARIMA(2,1,1) order was used to predict the imbalance variability index over the next 0.5 seconds. The forecast results showed that the imbalance variability index would continue to rise to 0.83, with a growth rate of 0.14 / second. Sensitivity analysis was used to identify the dominant imbalance factors, sequentially masking each level of metrics and observing the output changes. Results showed that center of mass trajectory parameters and plantar pressure distribution parameters had the greatest impact on imbalance assessment, with sensitivity coefficients of 0.31 and 0.28, respectively. Based on the results of the sensitivity analysis, the foot-strike imbalance variability index was recalculated using a weighted fusion method, resulting in a final value of 0.78 (normal range <0.3). Imbalance variability was mapped to a fall risk level using a risk classification method, with a four-level scale: safe (<0.3), warning (0.3-0.5), dangerous (0.5-0.7), and emergency (>0.7). The current foot-strike imbalance variability index of 0.78 is considered emergency, triggering the anti-fall vest's highest level of protection, activating the all-round airbag system.

[0032] Step S23 includes the following steps: Step S231: Calculating the center of gravity horizontal / vertical displacement change ratio based on the foot landing imbalance variability data to obtain the center of gravity horizontal / vertical displacement change ratio; Step S232: performing center of gravity trajectory swing speed increment analysis on the center of gravity horizontal / vertical displacement change ratio to generate center of gravity trajectory swing speed increment data; Step S233: Derivation of ankle joint inversion / valgus moment mutation based on the center of gravity horizontal / vertical displacement change ratio and the center of gravity trajectory swing speed increment data to obtain ankle joint inversion / valgus moment mutation data; Step S234: calculating the approximate mean difference of the torque increase distribution at adjacent time scales for the ankle joint inversion / valgus torque mutation data to generate the approximate mean difference of the torque increase distribution; Step S235: performing a tilt potential energy dynamic spatial superposition integral on the center of gravity trajectory swing velocity increment data and the approximate mean difference of the torque increase distribution based on the Hamiltonian integral to obtain tilt potential energy spatial superposition data; Step S236: performing tilt potential energy intensity increment analysis based on the tilt potential energy spatial superposition data, the center of gravity trajectory swing speed increment data, and the ankle joint inversion / valgus torque mutation data to obtain tilt potential energy intensity increment data.

[0033] In this embodiment of the present invention, in step S231, the data streams collected by the triaxial accelerometers and angular velocity gyroscopes located at the heel, mid-plantar, and forefoot regions of both feet in the inertial measurement unit are read to obtain the vertical acceleration extremes and ground reaction force directional changes at each moment of foot-ground contact during a continuous gait cycle. Using a 1 Hz cadence as a window, the vertical Z-axis acceleration integral for the entire stance phase from heel strike to forefoot lift-off within each gait cycle is extracted and recorded as the vertical displacement estimate. Simultaneously, the combined velocity curves for the X- and Y-axis directions within the same cycle are extracted and integrated. The combined velocity is obtained by taking the square root of the sum of the squares of the X- and Y-axis accelerations, and the integrated result is used as the horizontal displacement estimate. To reduce short-term posture sway interference, the horizontal and vertical displacements are denoised using a weighted sliding average window with a sliding window size of 0.5 seconds, a Gaussian distribution kernel as the weighting function, and a standard deviation of 0.2. After the filtering process is completed, the ratio of the change in the horizontal displacement of the center of gravity divided by the change in the vertical displacement is calculated for each step cycle, and recorded as the center of gravity horizontal / vertical displacement change ratio sequence. In the actual test, the gait data acquisition frequency is 50Hz, the total length of the test sample is 10 minutes, and a total of 600 data points are extracted for estimating the ratio for each gait cycle. The numerical range of this ratio is concentrated between 0.6 and 1.2 in normal gait. If there is an instantaneous jump exceeding 2.0 or falling below 0.4, it is marked as a center of gravity offset abnormal point. All ratio data and abnormal point labels are written into the central data buffer module as the input basis for subsequent steps. In step S232, the center of gravity horizontal / vertical displacement change ratio sequence extracted in step S231 is subjected to second-order difference processing, and the change between any two adjacent cycle ratios is calculated. Then, the change is subjected to time derivative processing to obtain the swing speed increment of the center of gravity trajectory per unit time. In the specific operation, the ratio sequence is aligned in time and sampled equidistantly, with each sample point interval of 0.2 seconds. Based on this, the ratio sequence is differentiated to obtain the difference between adjacent ratios. This difference is then divided by the time interval to obtain the velocity increment. The physical meaning of the velocity increment is the dynamic adjustment intensity caused by the proportional change in center of gravity displacement per unit time, reflecting the control required for ankle and knee fine-tuning. The velocity increment sequence is filtered with a third-order moving average filter with a filter window size of 0.6 seconds to prevent velocity abrupt changes caused by occasional sensor errors. The resulting center of gravity trajectory swing velocity increment data sequence is expressed in units of ratio change per second (dimensionless / second), which serves as an important dynamic indicator for subsequent torque abrupt change derivation. In actual gait testing, the center of gravity swing velocity increment is mainly concentrated between 0 and 0.9. When its peak exceeds 1.6, it is considered an abnormal control response. The maximum swing velocity increment in each cycle is stored in a one-to-one correspondence with the corresponding gait period number, forming a complete center of gravity swing dynamic database.In step S233, the center of gravity horizontal / vertical displacement change ratio obtained in step S231 and the center of gravity trajectory swing speed increment data obtained in step S232 are used to deduce the instantaneous torque mutation value of the inversion and eversion of the ankle joint through the dynamic lever arm derivation method. During the operation, the plantar contact surface and the ankle joint center position vector in each gait cycle are first constructed as a fixed lever arm in a two-dimensional plane. The length is set to 0.25 meters based on the physical parameters, and the direction is consistent with the opposite direction of the plantar acceleration vector. Then, the swing speed increment value is used as the acceleration change factor caused by the dynamic load and is introduced into the torque derivation formula. The specific torque value calculation method is: multiply the center of gravity horizontal / vertical displacement change ratio by the swing speed increment and multiply it by the lever arm length. The result is the inversion / eversion torque mutation value caused by the instantaneous force change of the ankle joint. The numerical unit is Newton-meter, and the accuracy is set to three decimal places. To determine the direction of torque mutation, inversion is set as the positive direction and eversion is set as the negative direction. Mutation points are then identified based on the torque change trend within each cycle. A mutation point is defined as the period when the torque change exceeds 2.8 Nm per unit time. The peak value, duration, and time of occurrence of the mutation are recorded together to form a data sequence of ankle inversion / eversion torque mutations. In sample tests, in more than 90% of normal gait cycles, the mutation value does not exceed 1.5 Nm and the duration is less than 0.3 seconds. If the mutation value exceeds 3.2 Nm and the duration is greater than 0.5 seconds, it is considered that there is a risk of control instability within that cycle. All ankle torque mutation data are bound to the labeled gait number, forming the basic mechanical data that can be used as a reference for buffer area mapping.

[0034] In step S234, the ankle inversion / valgus torque mutation data sequence recorded in step S233 is retrieved. Using each consecutive gait cycle as the basic time scale, a torque variation increase sequence between two adjacent cycles is constructed. This sequence is defined as the torque mutation peak value of the subsequent cycle minus the torque mutation peak value of the previous cycle. Positive values are included in the increasing trend sequence. Ten consecutive gait cycles are used as a sample window, and nine sets of adjacent torque mutation peak difference sequences are sequentially extracted. After averaging these nine torque increase value sequences, the absolute difference between each torque increase value and the mean is calculated. Finally, the average of all absolute differences is calculated to obtain the approximate mean difference of the torque increase distribution at this time scale. To improve the stability of the results, an overlapping window strategy is used when extracting each data window. Specifically, the starting point of the subsequent window slides backward by two gait cycles, forming a progressive sliding analysis mechanism with an 80% sample overlap. In actual sample analysis, when the cadence is stable at 1Hz, 600 samples are collected every 10 minutes, generating approximately 296 sets of approximate mean difference samples for the torque increase distribution. All mean difference values are uniformly expressed in Newton-meters, with three decimal places, and primarily range from 0.15 to 0.75. Groups with mean differences exceeding 1.25 are marked as abnormally elevated torque intervals. All approximate mean difference data for the torque increase distribution are stored in a structured array, with the original window number and corresponding timestamp appended, forming a continuous mechanical feature sequence for the dynamic potential energy deduction stage. In step S235, the center of gravity trajectory swing velocity increment data generated in step S232 and the torque increase distribution approximate mean difference data generated in step S234 are synchronized by timestamp to establish a one-to-one state time series set. Using each time node as the integral basis, a potential-kinetic energy coupling expression structure is constructed using the Hamiltonian transformation concept, where the center of gravity trajectory swing velocity increment is considered as the system kinetic energy weight term and the torque increase distribution approximate mean difference is considered as the system potential energy offset factor. The dynamic spatial energy state is defined as the superposition value of the product of the two indicators in the direction of the unit lever arm at each moment, and this value is cumulatively integrated at all sampling points. In the specific implementation, the gait data sampling frequency is 50Hz, and there are 3000 groups of data per minute. For each group, a corresponding superposition integral term is constructed, and the integral value is subjected to a moving integral accumulation operation in the time domain. The integration step is 20ms, and the integration window size is set to 500ms. The integral data within the window is convolved according to the weight function. The weight function uses a second-order Poisson kernel, the center coefficient is set to 1, the tail coefficient is reduced to 0.1, and the weight length does not exceed 25 points. All superposition integration operations are performed with fixed-point precision in the embedded processing unit, and the results are uniformly output as a sequence of tilted potential energy spatial superposition data in joules.This sequence represents a generalized potential energy representation derived from the integral superposition of dimensionless and moment quantities, representing the total energy state of the tilt dynamic trend per unit time and space. In actual gait samples, its value range is primarily concentrated between 12 and 85 joules, and data points exceeding 150 joules are typically accompanied by sharp fluctuations in displacement. This sequence data is written to the energy state database in groups by second, with each group accompanied by a gait index and sliding window number information. In step S236, the tilt potential energy spatial superposition data generated in step S235, the center of gravity trajectory swing velocity increment data from step S232, and the ankle joint inversion / valgus torque mutation data from step S233 are jointly analyzed to derive the tilt potential energy intensity increment at each time node. The operational process begins by performing first-order difference processing on the superimposed data sequence to obtain a difference sequence of the spatial variation of the tilt potential energy within a continuous time period, i.e., the potential energy variation trend per unit time. Each point in the change sequence is multiplied with its corresponding center of gravity swing velocity increment to form a tilt energy conversion increment weighted by velocity. This is then superimposed with the ankle joint torque mutation data at the corresponding moment to introduce an offset correction as a structural instability factor. After all these operations are performed uniformly, the result is the tilt potential energy intensity increment data, which physically represents the generalized tilt driving force change per unit time due to the combined effects of velocity excitation and structural torque adjustment. To ensure data stability, the sequence is normalized by standard deviation and then filtered by a three-point moving median filter with a filter window of no more than 0.1 seconds. In actual sampling, the tilt potential energy intensity increment unit is set to joules per second, with values mainly ranging from 2 to 12 joules per second. Any sudden increase exceeding 25 joules per second and lasting for more than 0.3 seconds is marked as an abnormal potential energy jump point in the data system. Finally, all tilt potential energy intensity increment data are uniformly stored in the characteristic state matrix, which serves as an important dynamic parameter basis for subsequent buffer trigger area decisions and attitude balance control responses.

[0035] Step S25 includes the following steps: Step S251: performing time-delay coupling characteristic analysis on the nonlinear coupling strength data between the joint muscle groups to obtain time-delay coupling characteristic data between the joint muscle groups; Step S252: performing tension-frequency-energy ratio analysis on the nonlinear coupling strength data between the joint muscle groups based on the time-delay coupling characteristic data between the joint muscle groups to obtain the nonlinear tension-frequency-energy ratio between the joint muscle groups; Step S253: performing muscle group stretching and deviation correction reaction simulation analysis on the nonlinear tension-frequency-energy ratio between each joint muscle group to obtain muscle group stretching and deviation correction reaction data; Step S254: performing muscle group-potential energy falling posture tendency coupling analysis based on the muscle group stretching correction reaction data and the tilt potential energy intensity increment data to obtain muscle group-potential energy falling posture tendency coupling data; Step S255: performing a fall posture tendency simulation deduction based on the muscle group-potential energy fall posture tendency coupling data to obtain fall posture tendency deduction data.

[0036] In an embodiment of the present invention, first, a multi-channel surface electromyography acquisition system is used to perform continuous dynamic posture monitoring on the test subject of the wearable elderly care anti-fall vest. The sampling frequency is set to 1000 Hz, and the acquisition objects include the electrical signals of the main control muscle groups at the hip joint, knee joint and ankle joint. The specific muscle groups include the gluteus maximus, quadriceps femoris, hamstrings, gastrocnemius and tibialis anterior. The acquisition time period is set to 20 seconds of continuous walking. The collected original electrical signals are high-pass filtered, and the filter lower limit is set to 20 Hz to remove motion artifacts. Then, low-pass filtering is performed, and the filter upper limit is 450 Hz to remove high-frequency noise interference. The signals after the above filtering process are normalized respectively and standardized to a sequence with a mean of zero and a variance of one. Subsequently, pairs of electrical signal sequences between muscle groups were selected. Based on the delayed mutual information analysis method, the delay step range was set from 0 to 50 milliseconds. The function graph of the mutual information value as a function of the delay time was calculated. By extracting the time delay corresponding to the mutual information maximum point, the dominant coupling delay between the muscle group pairs was determined, and a time-delay coupling characteristic matrix containing ten muscle group pairs was constructed. Each cell in this matrix represents a millisecond-level delay value, and the matrix dimensions are 5 rows and 5 columns (excluding the diagonal). This matrix serves as the input data for step S252. The time-delay coupling characteristic matrix obtained in step S251 is called and timing calibration is performed for each muscle group pair. This means that the original electrical signal sequence is shifted forward or backward according to the corresponding delay value to form a synchronous coupling input sequence. The synchronized signal is subjected to spectral analysis using a short-time Fourier transform, with the window function type set to a Henning window, the window length set to 1024 points, and the sliding step size set to 256 points. The power spectral density in the transformed result, within the range of 0 to 50 Hz, extracts the integrated energy values of the low-frequency band (0 to 10 Hz) and the mid-frequency band (10 to 30 Hz), recorded as the low-frequency energy value and the mid-frequency energy value, respectively. The tension-frequency energy ratio is then calculated for each muscle group pair by dividing the mid-frequency energy value by the low-frequency energy value to obtain the nonlinear tension-frequency energy ratio data. This ratio indicates the degree of tension control deviation caused by coupling hysteresis in this muscle group pair during dynamic gait. For example, in one experimental subject, the tension-frequency energy ratio between the quadriceps femoris and gastrocnemius muscles was 2.3, indicating that the tension in the mid-frequency band was significantly higher than that in the low-frequency band, suggesting that the coupling relationship tends to be dominated by a rapid contraction response. Across 20 experimental samples, the tension-frequency energy ratios for each muscle group pair ranged from 0.85 to 3.1, showing significant individual variability and serving as the basis for subsequent correction simulations. Based on the nonlinear tension-frequency energy ratio data obtained in step S252, the tensile correction response of the joint muscle group under external force disturbance was simulated. The standard elastic-viscous response framework of the tendon-muscle unit was employed, with the tension-frequency-energy ratio as the modulator of the intrinsic activation rate. The simulation started with a passive 10 mm stretch of the tendon, corresponding to a 15% increase in the physiological muscle length. Initially, the active contraction mechanism was inactive. The intrinsic tension feedback path after muscle stretch was iteratively simulated using the finite difference method, with a time step of 0.002 seconds and a total simulation duration of 2 seconds.At each time step, the muscle group's reverse force response is iteratively updated based on the current stretch level and the tension activation correction value corresponding to the tension-frequency-energy ratio. Specifically, when high-frequency dominance is set (i.e., the tension-frequency-energy ratio is greater than 2.0), the activation response time is advanced by 0.15 seconds, and the peak tension is increased by 20%. When low-frequency dominance is set (i.e., the tension-frequency-energy ratio is less than 1.0), the activation response is delayed by 0.3 seconds, and the peak tension is reduced by 15%. The simulation outputs a time series of stretch-correction reaction forces for each muscle group, recording metrics such as maximum reaction force, reaction time, and tension release slope. Among the hip joint muscles, the hamstrings achieve a peak correction force of 14.2 Newtons, occurring at 0.48 seconds into the simulation, demonstrating their strong ability to generate active tension for correction. However, the gastrocnemius' peak correction response is only 9.1 Newtons, occurring at 0.76 seconds, indicating a slower response. These muscle group stretch-correction reaction data serve as input for subsequent posture-inclined coupling analysis.

[0037] First, the muscle group stretch correction response data obtained in step S253 are called. Specifically, they include the active tension response time series, peak reaction force value, slope increment, and hysteresis time of each major joint muscle group after passive stretching during the standardized gait process. At the same time, the tilt potential energy intensity increment data obtained in step S24 are called and introduced. This data is the rate of change of potential energy per unit time, which is derived from the body forward lean angular velocity, center of mass height change rate, and gait deviation angle obtained by the inertial measurement unit, in joules per second. Based on these two types of data, a multidimensional time series coupling analysis method is adopted. Under a unified time index, the muscle group stretch correction response signal and the tilt potential energy intensity increment signal are subjected to cross-covariance sliding window analysis. The window width is set to 200 milliseconds and the sliding step size is set to 20 milliseconds. The covariance coefficient sequence in each time window is calculated throughout the entire continuous gait cycle, and the key coupling points are extracted by the positive and negative polarity changes and absolute value of the covariance coefficient. To ensure directional consistency in the analysis, all data were first time-aligned. This means that the peak response of the corrective reaction was strictly aligned with the moment of the tilt potential energy mutation, with an error tolerance of less than 10 milliseconds. This was then used to calculate the coupling delay error distribution, which measures the time distribution of when the potential energy mutation preceded or lagged behind the muscle group's response. Statistical results showed that in the coupling relationship between the hamstrings and hip potential energy, the potential energy mutation lagged the muscle group's corrective response by an average of approximately 70 milliseconds, and the average covariance of the coupling strength was 0.81, indicating a strong temporal synchronization between this muscle group and the tilt potential energy. The covariance time series curves of all coupling pairs were normalized to the interval [-1, 1]. The coupling pattern sequence was extracted based on the covariance peak position and width. A 5×3-dimensional muscle group-potential energy fall posture tendency coupling data matrix was generated, corresponding to the muscle group number and coupling pattern. Rows represent muscle groups, and columns represent the three potential energy directions: forward lean, side lean, and backward lean. Each cell was populated with the maximum covariance value and lag time of the coupling pattern to construct a coupling feature vector, which was used as input for the next step of fall posture tendency deduction. The muscle group-potential energy fall posture tendency coupling data matrix generated in step S254 was used as the deduction basis for the stepwise evolution of the fall state using a rule-driven state evolution algorithm. This deduction does not use a data-driven model, but instead employs a set of state discrimination rules based on physical principles and human biomechanical behavior. The specific deduction logic was set as follows: the initial posture state was a standing stable state, the center of mass height was recorded as 0.91 meters, the forward lean angle was set to 0 degrees, and the gait period was 1 second. In each time step, the change value of posture potential energy per unit time is derived based on the incremental value of tilt potential energy, and the current muscle group correction reaction data is compared to see whether sufficient reaction tension can be generated within the specified lag time to offset the incremental value of tilt potential energy.The derivation logic is as follows: If the current tilt potential energy increment exceeds the maximum tension compensation power of the muscle group (calculated by multiplying the peak tension of the correction by the applied displacement length and dividing it by the response time) within the muscle group's delayed response time, the subject is marked as prone to imbalance. If the subject remains in a prone to imbalance state for three consecutive time steps and the total center of gravity offset exceeds a radius of 10 cm from the support boundary, the subject is marked as initiating a fall posture. Based on this state progression rule, derivation is performed separately for each posture direction. For example, under the conditions of a hamstring delayed response time of 110 milliseconds, a peak tension of 12.8 Newtons, and a corrective displacement of 6 mm, the tension compensation power per unit time is 0.7 watts. When the forward tilt potential energy increment reaches 1.2 watts and is maintained for 0.16 seconds, the subject is marked as initiating a forward tilt posture fall. This state derivation process is repeated, ultimately generating derivation data such as the time point of fall tendency, posture angle, and imbalance duration for each subject in the three posture directions. This data is output as a structured data sequence for subsequent control response design and evaluation.

[0038] Step S254 includes the following steps: The muscle group stretch correction response data is decomposed into dynamic tension gradients to obtain the axial tension component and radial tension component of each joint muscle group; Perform multi-dimensional plane projection decomposition on the tilt potential energy intensity increment data to obtain multi-dimensional tilt potential energy component data; Based on the multi-dimensional tilt potential energy component data, the nonlinear gradient change analysis of the human body tilt angle and angular acceleration is performed to generate the human body tilt angle change data and tilt angular acceleration change data respectively; The muscle group-potential energy falling posture tendency coupling analysis was performed based on the human body tilt angle change data, tilt angle acceleration change data and the axial tension component and radial tension component of each joint muscle group to obtain the muscle group-potential energy falling posture tendency coupling data.

[0039] In an embodiment of the present invention, in the process of decomposing the dynamic tension gradient of the muscle group stretch correction reaction data, the joint position-time series data recorded in the inertial measurement unit and the muscle group electrical activity signal recorded by the surface electromyography sensor are first synchronized and processed, and all data are time-aligned according to the sampling frequency of 1000 Hz, and the error threshold is controlled within 1 millisecond. For each major joint (such as the hip joint, knee joint and ankle joint), the displacement data of the main axis direction and the vertical direction during the gait cycle are extracted, and the amplitude change of the electromyography signal is combined to establish a tension change gradient map. The central difference method is used to perform first-order derivative processing on 5 points before and after any moment in the time series to obtain the gradient data of tension changing with time. According to the skeletal reference coordinate system of the joint, the tension vector of each muscle group is decomposed into an axial tension component along the skeletal direction and a radial tension component perpendicular to the skeletal direction. In specific implementation, taking the role of the hamstrings in knee flexion and extension as an example, the principal axis of knee flexion and extension was set as the axial direction. The hamstring tension gradient reached its maximum value at 0.85 seconds, with a gradient of 11.3 Newtons per second. Using the vector projection function, the axial tension component of the hamstrings at this moment was determined to be 9.6 Newtons per second, and the radial tension component was 4.5 Newtons per second. The tension changes of all joint muscle groups were similarly decomposed, and the axial and radial tension sequences of each muscle group throughout the gait cycle were output as the basis for subsequent analysis. In the multidimensional planar projection decomposition step of the tilt potential energy intensity increment data, the instantaneous acceleration and angular velocity signals in the X (front-back), Y (left-right), and Z (up-down) directions were first extracted based on the accelerometer and gyroscope data in the six-axis inertial measurement unit. The sampling frequency was 500Hz, and a third-order Savitzky-Golay filter was used for filtering, with a window width of 9. Using the body's center of mass as the reference point, a three-dimensional acceleration vector is constructed using triaxial acceleration data. Dot products are then performed with the unit vector in each tilt direction to obtain the projected potential energy change in each plane at the current time point. The projection directions are set to the XY plane (horizontal), YZ plane (lateral tilt), and XZ plane (forward tilt). For example, a set of data recorded during an actual experiment shows that at 0.92 seconds into the gait cycle, the forward acceleration is 2.3 meters per square second, the left and right accelerations are 0.9 meters per square second, and the vertical acceleration is -0.4 meters per square second. Combined with a human mass distribution model (using 72 kg as an example), the unit potential energy increments in the forward tilt direction are 1.68 joules, 0.83 joules in the lateral tilt direction, and -0.28 joules in the vertical direction. This results in a multi-dimensional tilt potential energy component data sequence, with each time point containing scalar potential energy increments in the three directions.

[0040] In the process of analyzing the nonlinear gradient changes in human body tilt angle and angular acceleration based on multi-dimensional tilt potential energy component data, the tilt potential energy change sequence in the YZ and XZ planes is first nonlinearly interpolated using cubic spline interpolation to improve the response sensitivity to extreme points. The smoothed potential energy curve obtained by interpolation is then processed by second-order derivative to obtain an approximate trend of angular acceleration. In the angle analysis, the angular velocity integral data of the inertial measurement unit is combined with the gradient fitting function to extract the angle change speed, and the tilt angle change value is further accumulated and integrated. Taking the roll direction as an example, in the test data, the roll angular velocity of a certain subject increased from 0.72 radians per second to 1.44 radians per second between 1.16 seconds and 1.18 seconds, corresponding to an increase in the roll angle from 6.2 degrees to 11.3 degrees. During this period, the maximum angular acceleration reached 7.1 radians per square second. The synchronization between this data and the potential energy component change in the roll direction was greater than 0.92. Using the minimum sum of squared error residuals criterion for fitting, the error was controlled within 0.08 radians, ultimately forming a time series of tilt angle change data and tilt angle acceleration change data. In the step of conducting a muscle group-potential energy fall posture tendency coupling analysis based on the human body's tilt angle change data, tilt angle acceleration change data, and the axial and radial tension components of each joint muscle group, the tilt angle data and angular acceleration data were first used as driving variables, and the axial and radial tension components as response variables. A bivariate lagged cross-correlation analysis method was used to analyze the correlation between the two data sets within a sliding range of a maximum lag time of 300 milliseconds. For each time window, the cross-correlation coefficient between the driving and response variables within that window and the lag time corresponding to the maximum value were calculated. For the hip flexor muscles and anteversion angular acceleration, for example, the maximum cross-correlation coefficient was 0.89, with a lag time of 90 milliseconds, between 1.0 and 1.2 seconds, indicating a strong temporal response coupling between this muscle group and changes in anteversion posture. This process was repeated for all major muscle groups and posture data in three directions, constructing a 9×3 coupling matrix. Rows represent muscle group numbers, columns represent tilt directions, and each cell contains a coupling vector consisting of the maximum cross-correlation value and the lag time. This output, structured muscle group-potential energy fall posture tendency coupling data, served as input for subsequent simulations.

[0041] Step S3 includes the following steps: Step S31: performing convolution processing on the fall posture tendency deduction data to obtain fall posture tendency convolution data; Step S32: normalizing the tilt potential energy intensity increment data to obtain tilt potential energy intensity increment normalized data; Step S33: performing partitioned buffer expansion control on the anti-fall vest according to the convolution data of the falling posture tendency and the normalized data of the tilt potential energy intensity increment to obtain partitioned buffer expansion control data; Step S34: performing automated firmware design based on the partition buffer expansion control data to obtain partition buffer expansion control firmware.

[0042] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes: Step S31: performing convolution processing on the fall posture tendency deduction data to obtain fall posture tendency convolution data; In an embodiment of the present invention, the obtained fall posture tendency deduction data is used as the input tensor, and a one-dimensional temporal convolution processing method is used to extract the mutation gradient features therein. The convolution kernel width is set to 5, the step size is set to 1, and the time dimension is processed in an unfilled manner. The ReLU activation function is used to retain the positive mutation features while eliminating the influence of low-frequency redundant data. The processing flow constructs a three-layer convolution channel, with the first layer having 16 output channels, the second layer having 32 channels, and the third layer having 64 channels. Batch normalization operations are introduced after each layer of convolution to suppress feature drift and avoid feature distortion caused by differences in the dynamic gradient amplitude of the deduced data over a long period of time. After the convolution processing is completed, the posture tendency data of all time segments are mapped into a 128-dimensional posture tendency convolution vector through a global average pooling method, which serves as the high-dimensional feature input basis for partition buffer expansion control. A total of 223 groups of falling posture tendency deduction samples were collected in the actual experiment. The time length of each group of data was 10 seconds, and the sampling rate was 50Hz per second. After convolution processing, an average of 43.7 high-order feature points in the tendency mutation gradient sequence were extracted as the data input basis for subsequent steps.

[0043] Step S32: normalizing the tilt potential energy intensity increment data to obtain tilt potential energy intensity increment normalized data; In an embodiment of the present invention, the obtained tilt potential energy intensity increment data is normalized to eliminate the data scale deviation caused by the sensor acquisition range and terrain differences. Specifically, a maximum-minimum linear normalization algorithm is used to map each set of tilt potential energy increment values to an interval between 0 and 1. The normalization expression is: the current data minus the minimum value and then divided by the difference between the maximum and minimum values. In the actual experiment, inertial measurement data in the horizontal, longitudinal, and vertical directions are selected to construct three-dimensional tilt potential energy intensity tensors, each dimension with a length of 500 time steps, corresponding to 10 seconds of data. Through the normalization operation, the original maximum tilt potential energy increment amplitude difference in each dimension is compressed from 2.7N·m to 14.6N·m into a standardized vector, so that the potential fall dynamics under subsequent changes in body posture have a unified dimensional standard, preventing the tensor characteristics of a certain direction from causing unexpected interference to the control strategy weight. After normalization, all tensors are uniformly converted to 32-bit floating point format for storage for further partition buffer decision processing in step S33.

[0044] Step S33: performing partitioned buffer expansion control on the anti-fall vest according to the convolution data of the falling posture tendency and the normalized data of the tilt potential energy intensity increment to obtain partitioned buffer expansion control data; In an embodiment of the present invention, the fall posture tendency convolution vector obtained in step S31 is fused with the normalized tilt potential energy intensity increment tensor in step S32, and a dual-input channel mapping mechanism is used to input data from two different sources into a three-layer neural tensor matching network. The first layer compresses the 128-dimensional convolution feature vector to 64 dimensions through linear mapping, and processes the tilt potential energy data with the same structure. Then, a cross-channel product operation is introduced in the second layer to extract the response gain matrix between the tendency posture and the potential energy evolution. This gain matrix is the sensitive response coefficient of the body posture change to the local gravitational potential energy disturbance at each time step. The third layer uses this matrix to generate a partitioned buffer response strategy. During the buffer control mapping, the vest area is divided into five major buffer units: shoulder, back, left waist, right waist, and buttocks. The buffer airbag pressure control coefficients are set to 0.8, 1.0, 1.1, 1.1, and 1.3 respectively. According to the output value of the fused response matrix, the expansion delay time and target pressure intensity of each buffer are adjusted through the mapping strategy. Taking the 94th set of data from the experimental sample as an example, when it is detected that the right-side tilt angle acceleration shows a significant nonlinear increase and the waist tension gradient rises by more than 18%, the system sets the activation delay of the right waist and hip buffer unit to 0.3 seconds, the target pressure to 0.42MPa, and generates a partitioned buffer expansion control data instruction set.

[0045] Step S34: performing automated firmware design based on the partition buffer expansion control data to obtain partition buffer expansion control firmware.

[0046] In this embodiment of the present invention, the logical instruction arrangement of automated firmware is designed based on the partitioned buffer expansion control data instruction set output in step S33. This firmware is written in C language using an embedded architecture, with a set main frequency of 72 MHz and based on a 32-bit ARM Cortex-M4 core architecture. Each partitioned buffer module is bound to a set of independently executable instruction sequences, which contain three types of control instructions: expansion delay setting instructions, target pressure adjustment instructions, and termination threshold determination instructions. These control instructions trigger specific solenoid valve opening and closing actions using GPIO interrupt mode. The solenoid valve response time is set to less than 25 ms, and the air pump pressure adjustment frequency is set to 0.1 Hz to reduce current fluctuation interference. In the buffer airbag pressure monitoring section, a pressure sensor compares feedback data with the target pressure data in real time. Error determination is performed through the ADC channel, and a stop signal is automatically transmitted when the error falls below 0.03 MPa. The firmware is ultimately packaged in hexadecimal format for future feature correction and system parameter tuning.

[0047] Step S33 includes the following steps: Step S331: performing wavelet transform on the convolution data of the falling posture tendency to obtain posture tendency frequency domain feature data; Step S332: determining the vest partition expansion priority sequence based on the posture tendency frequency domain feature data to obtain the vest partition expansion priority sequence; Step S333: performing a calculation for the partitioned airbag inflation pressure gradient distribution based on the vest partition inflation priority sequence and the tilt potential energy intensity increment normalization data to obtain a partitioned airbag pressure control parameter set; Step S334: performing partitioned cushioning expansion control of the anti-fall vest based on the partitioned airbag pressure control parameter set to obtain partitioned cushioning expansion control data.

[0048] In this embodiment of the present invention, the input fall posture tendency convolution data is subjected to a three-layer discrete wavelet decomposition along the time series dimension. Processing is performed using the Daubechies-4 wavelet basis function, with a sampling frequency of 100 Hz and a wavelet decomposition level of 3. High-frequency coefficient groups (D1, D2, D3) and low-frequency coefficient groups (A3) are extracted. The high-frequency coefficients are used to extract sudden posture change trends, while the low-frequency coefficients are used to extract gradual posture change trends. In practice, for each set of fall posture tendency convolution data matrix (128×64 in size, where 128 represents the time step and 64 represents the sensing channel), a wavelet transform is performed on each column of channel data along the time axis. The resulting wavelet coefficient matrix is subjected to energy spectral density analysis and main frequency band statistics to extract the main response frequency band between 0.8 Hz and 3.2 Hz. This is then subjected to energy weight normalization to output a posture tendency frequency domain feature data matrix. The matrix size is maintained at 64×3, with each column representing the relative energy distribution of each channel in the main frequency band. After receiving the posture tendency frequency domain feature data output in step S331, the 64 signals are distributed to 10 regions on the vest (left and right shoulders, left and right waist, left and right back, left and right hips, front chest, and back) based on the channel-specific vest physical partition mapping table. Each region is mapped to an average of 6 to 7 signal channels. A partition energy priority vector (length 10) is generated by performing mean filtering on the channel frequency domain energy values within each region and sorting them from high to low. During the sorting process, a threshold of 0.05 is introduced to merge adjacent energies to avoid misjudgment caused by overlapping spectral responses at the boundaries of multiple regions. The vest partition expansion priority sequence is output in descending order of partition energy priority. The structure is a one-dimensional integer vector of length 10 with a value range of [1,10], where 1 represents the highest priority and 10 represents the lowest. Based on the vest zone inflation priority sequence generated in step S332 and the normalized data for the tilt potential energy intensity increment generated in step S32 (set to the normalized value x, in the range [0,1]), the zone airbag inflation pressure gradient distribution is calculated. A base inflation pressure of 40 kPa is set for each vest zone. Based on this, a weighting factor is assigned based on the inflation priority sequence number: 1 for priority 1, 1.5 for priority 10, 0.5 for priority 10, and linearly decreasing for the remaining zones. Specifically, the weighting factor is equal to (1.5 - 0.1 × (priority - 1)). The target pressure for each zone is the base pressure multiplied by the weighting factor, then multiplied by the normalized value x. For example, if x = 0.8, the final target pressure for the zone with priority 1 is 40 × 1.5 × 0.8 = 48 kPa, while the target pressure for the zone with priority 10 is 40 × 0.5 × 0.8 = 16 kPa. All calculation results are encapsulated into a partitioned airbag pressure control parameter set, which is a one-dimensional floating-point vector with a length of 10, corresponding to the airbag inflation target pressure of each vest partition.

[0049] Based on the zoned airbag pressure control parameter set generated in step S333, the vest's zoned airbags are inflated in parallel. The controller sampling frequency is set to 10 Hz, and a closed-loop PID control algorithm is used for real-time pressure regulation. Each zoned airbag is equipped with an independent piezoelectric film pressure sensor and micro-solenoid valve. The sensor measurement error does not exceed ±0.2 kPa, and the feedback cycle is 100 milliseconds. The control logic is set according to the target pressure corresponding to the pressure control parameter set. The air pump flow rate is controlled by adjusting the PWM duty cycle, with the initial duty cycle set to 70%. During the inflation process, deviation analysis is performed based on the sensor feedback pressure, and the air pump drive signal is adjusted in real time until the pressure value remains stable within an error range of ±0.5 kPa. The control delay of the entire inflation process does not exceed 500 milliseconds. The final zoned buffer inflation control data output consists of 10 sets of pressure sampling points, each set containing 5 stable point samples, which are used for subsequent firmware mapping and execution data package construction. This control data serves as the basis for the final buffering decision of the anti-fall control link and is written into the control firmware for the hardware layer to execute inflation commands.

[0050] The present invention also provides a control optimization system for an elderly care anti-fall vest, which is used to execute the above-mentioned control optimization method for an elderly care anti-fall vest. The control optimization system for an elderly care anti-fall vest comprises: The data selection module is used to obtain the data set of the elderly's falling behavior status and the data set of muscle stress tension fluctuation records reported by users; perform feature selection on the data set of the falling behavior status and the data set of muscle stress tension fluctuation records to obtain the data set of the falling behavior status feature selection and the data set of the muscle stress tension fluctuation record feature selection; The fall posture tendency simulation module is used to deduce the foot landing imbalance variability based on the data set selected from the fall behavior state characteristics to obtain foot landing imbalance variability data; perform tilt potential energy intensity increment analysis based on the foot landing imbalance variability data to obtain tilt potential energy intensity increment data; and perform fall posture tendency simulation and deduction based on the data set selected from the muscle stress tension fluctuation record characteristics and the tilt potential energy intensity increment data to obtain fall posture tendency deduction data; The partition buffer expansion control module is used to control the partition buffer expansion of the anti-fall vest according to the falling posture tendency deduction data to obtain the partition buffer expansion control data; and to perform automated firmware design based on the partition buffer expansion control data to obtain the partition buffer expansion control firmware.

[0051] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A control optimization method for an anti-fall vest for elderly care, characterized in that: The following steps are involved: Step S1: Obtaining a dataset of the elderly person's falling behavior status and a dataset of muscle stress tension fluctuation records based on user feedback; Performing feature selection on the falling behavior state dataset and the muscle stress tension fluctuation record dataset to obtain a falling behavior state feature selection dataset and a muscle stress tension fluctuation record feature selection dataset; Step S2: performing foot landing imbalance variability deduction on the selected dataset of fall behavior state features to obtain foot landing imbalance variability data; performing tilt potential energy intensity increment analysis based on the foot landing imbalance variability data to obtain tilt potential energy intensity increment data; performing fall posture tendency simulation deduction based on the selected dataset of muscle stress tension fluctuation records and the tilt potential energy intensity increment data to obtain fall posture tendency deduction data; Step S3: performing partition buffer expansion control of the anti-fall vest according to the fall posture tendency deduction data to obtain partition buffer expansion control data; performing automated firmware design based on the partition buffer expansion control data to obtain partition buffer expansion control firmware.

2. The control optimization method for the elderly care anti-fall vest according to claim 1 is characterized in that: Step S2 includes the following steps: Step S21: performing a pre-fall cadence structure analysis on the fall behavior state feature selection data set to obtain pre-fall cadence structure data; Step S22: performing foot landing imbalance variability deduction on the fall behavior state feature selection data set based on the pre-fall cadence structure data to obtain foot landing imbalance variability data; Step S23: performing tilt potential energy intensity increment analysis based on the foot landing imbalance variability data to obtain tilt potential energy intensity increment data; Step S24: performing nonlinear coupling strength analysis between joint muscle groups on the muscle stress tension fluctuation record feature selection data set to obtain nonlinear coupling strength data between joint muscle groups; Step S25: performing a fall posture tendency simulation deduction based on the nonlinear coupling strength data between the joint muscle groups and the tilt potential energy strength increment data to obtain fall posture tendency deduction data.

3. The control optimization method for the elderly care anti-fall vest according to claim 2 is characterized in that: Step S22 includes the following steps: Step S221: performing a cadence acceleration fluctuation analysis on the cadence structure data before the fall to obtain cadence acceleration fluctuation data; Step S222: performing disordered single / double-foot support time ratio identification on the cadence structure data before the fall based on the cadence acceleration fluctuation data to obtain disordered single / double-foot support time ratio data; Step S223: performing support base impairment assessment on the disordered data of single / double-foot support time ratio to generate support base impairment data; Step S224: Deducing the angle difference of the center of gravity disordered offset trajectory based on the support base loss data and the disordered data of the single / double-foot support time ratio to obtain the angle difference of the center of gravity disordered offset trajectory; Step S225: Derivation of the pressure distribution offset of the foot contact surface based on the angle difference of the center of gravity disordered offset trajectory to obtain the pressure distribution offset data of the foot contact surface; Step S226: Deducing the foot landing imbalance variability based on the angle difference of the center of gravity disordered offset trajectory and the pressure distribution offset data of the foot landing contact surface to obtain the foot landing imbalance variability data.

4. The control optimization method for the elderly care anti-fall vest according to claim 2 is characterized in that: Step S23 includes the following steps: Step S231: Calculating the center of gravity horizontal / vertical displacement change ratio based on the foot landing imbalance variability data to obtain the center of gravity horizontal / vertical displacement change ratio; Step S232: performing center of gravity trajectory swing speed increment analysis on the center of gravity horizontal / vertical displacement change ratio to generate center of gravity trajectory swing speed increment data; Step S233: Derivation of ankle joint inversion / valgus moment mutation based on the center of gravity horizontal / vertical displacement change ratio and the center of gravity trajectory swing speed increment data to obtain ankle joint inversion / valgus moment mutation data; Step S234: calculating the approximate mean difference of the torque increase distribution at adjacent time scales for the ankle joint inversion / valgus torque mutation data to generate the approximate mean difference of the torque increase distribution; Step S235: performing a tilt potential energy dynamic spatial superposition integral on the center of gravity trajectory swing velocity increment data and the approximate mean difference of the torque increase distribution based on the Hamiltonian integral to obtain tilt potential energy spatial superposition data; Step S236: performing tilt potential energy intensity increment analysis based on the tilt potential energy spatial superposition data, the center of gravity trajectory swing speed increment data, and the ankle joint inversion / valgus torque mutation data to obtain tilt potential energy intensity increment data.

5. The control optimization method for the elderly care anti-fall vest according to claim 2 is characterized in that: Step S25 includes the following steps: Step S251: performing time-delay coupling characteristic analysis on the nonlinear coupling strength data between the joint muscle groups to obtain time-delay coupling characteristic data between the joint muscle groups; Step S252: performing tension-frequency-energy ratio analysis on the nonlinear coupling strength data between the joint muscle groups based on the time-delay coupling characteristic data between the joint muscle groups to obtain the nonlinear tension-frequency-energy ratio between the joint muscle groups; Step S253: performing muscle group stretching and deviation correction reaction simulation analysis on the nonlinear tension-frequency-energy ratio between each joint muscle group to obtain muscle group stretching and deviation correction reaction data; Step S254: performing muscle group-potential energy falling posture tendency coupling analysis based on the muscle group stretching correction reaction data and the tilt potential energy intensity increment data to obtain muscle group-potential energy falling posture tendency coupling data; Step S255: performing a fall posture tendency simulation deduction based on the muscle group-potential energy fall posture tendency coupling data to obtain fall posture tendency deduction data.

6. The control optimization method for the elderly care anti-fall vest according to claim 5 is characterized in that: Step S254 includes the following steps: The muscle group stretch correction response data is decomposed into dynamic tension gradients to obtain the axial tension component and radial tension component of each joint muscle group; Perform multi-dimensional plane projection decomposition on the tilt potential energy intensity increment data to obtain multi-dimensional tilt potential energy component data; Based on the multi-dimensional tilt potential energy component data, the nonlinear gradient change analysis of the human body tilt angle and angular acceleration is performed to generate the human body tilt angle change data and tilt angular acceleration change data respectively; The muscle group-potential energy falling posture tendency coupling analysis was performed based on the human body tilt angle change data, tilt angle acceleration change data and the axial tension component and radial tension component of each joint muscle group to obtain the muscle group-potential energy falling posture tendency coupling data.

7. The control optimization method for the elderly care anti-fall vest according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31: performing convolution processing on the fall posture tendency deduction data to obtain fall posture tendency convolution data; Step S32: normalizing the tilt potential energy intensity increment data to obtain tilt potential energy intensity increment normalized data; Step S33: performing partitioned buffer expansion control on the anti-fall vest according to the convolution data of the falling posture tendency and the normalized data of the tilt potential energy intensity increment to obtain partitioned buffer expansion control data; Step S34: performing automated firmware design based on the partition buffer expansion control data to obtain partition buffer expansion control firmware.

8. The control optimization method for the elderly care anti-fall vest according to claim 7 is characterized in that: Step S33 includes the following steps: Step S331: performing wavelet transform on the convolution data of the falling posture tendency to obtain posture tendency frequency domain feature data; Step S332: determining the vest partition expansion priority sequence based on the posture tendency frequency domain feature data to obtain the vest partition expansion priority sequence; Step S333: performing a calculation for the partitioned airbag inflation pressure gradient distribution based on the vest partition inflation priority sequence and the tilt potential energy intensity increment normalization data to obtain a partitioned airbag pressure control parameter set; Step S334: performing partitioned cushioning expansion control of the anti-fall vest based on the partitioned airbag pressure control parameter set to obtain partitioned cushioning expansion control data.

9. A control optimization system for elderly care anti-fall vest, characterized in that: For executing the elderly care anti-fall vest control optimization method according to claim 1, the elderly care anti-fall vest control optimization system comprises: The data selection module is used to obtain the data set of the elderly's falling behavior status and the data set of muscle stress tension fluctuation records reported by users; perform feature selection on the data set of the falling behavior status and the data set of muscle stress tension fluctuation records to obtain the data set of the falling behavior status feature selection and the data set of the muscle stress tension fluctuation record feature selection; The fall posture tendency simulation module is used to deduce the foot landing imbalance variability based on the data set selected from the fall behavior state characteristics to obtain foot landing imbalance variability data; perform tilt potential energy intensity increment analysis based on the foot landing imbalance variability data to obtain tilt potential energy intensity increment data; and perform fall posture tendency simulation and deduction based on the data set selected from the muscle stress tension fluctuation record characteristics and the tilt potential energy intensity increment data to obtain fall posture tendency deduction data; The partition buffer expansion control module is used to control the partition buffer expansion of the anti-fall vest according to the falling posture tendency deduction data to obtain the partition buffer expansion control data; and to perform automated firmware design based on the partition buffer expansion control data to obtain the partition buffer expansion control firmware.

Images