Dementia patient adaptive care decision system based on dynamic demand profiling

Abstract

This invention belongs to the field of medical and health technology. It discloses an adaptive care decision-making system for elderly people with dementia based on dynamic needs profiling. The system includes: collecting multi-source vital signs and spatial location signals; extracting physiological rhythms, cognitive decline, behavioral trajectories, and emotional fluctuations after data cleaning to construct a multi-layered needs profile; acquiring real-time monitoring data; performing dynamic state identification and causal analysis; selecting and personalizing suitable solutions from a care strategy library based on the causal analysis results; generating adaptive care decisions; and distributing these decisions to multiple terminals for storage and collaborative interaction. This significantly improves the efficiency of care for elderly people with dementia.

Description

Technical Field

[0001] This invention relates to the field of medical and health technology, and more specifically, to an adaptive care decision-making system for elderly people with dementia based on dynamic needs profiling. Background Technology

[0002] As the aging population continues to deepen, the care needs of the elderly with dementia have become a significant challenge for the field of healthcare information technology. Dementia patients exhibit complex symptoms such as memory impairment, disorientation, behavioral abnormalities, and mood swings due to progressive cognitive decline, resulting in highly individualized and dynamic care needs. In recent years, wearable health monitoring devices, remote care platforms, and other information technologies have been widely used in elderly health management, providing caregivers with basic status monitoring and abnormality warning functions through sensor data collection and information processing.

[0003] Existing dementia care monitoring systems suffer from significant problems in personalized needs identification and adaptive care decision-making. Specifically, the cognitive function of dementia patients continues to decline, their circadian rhythms become disordered, their behavioral patterns shift, and their emotional responses fluctuate irregularly. Furthermore, the differences in the physiological reserves and developmental trajectories corresponding to these functions demonstrate the individual heterogeneity and dynamic fluctuations in the state of dementia patients. This complex dynamic nature of individual characteristics makes it difficult for traditional monitoring methods based on fixed thresholds to accurately identify individualized abnormal states. Existing systems typically use uniform standardized parameters for anomaly detection, lacking the technical capability to transform multi-dimensional long-term monitoring data into individual needs profiles. They also cannot iteratively update these profiles in real time based on the dynamic evolution of the elderly person's condition, ultimately leading to a systematic deviation between care responses and actual individual needs. This technical limitation, lacking personalized dynamic modeling and adaptive decision-making capabilities, limits the system to simple, passive threshold alarms, preventing it from proactively generating targeted care intervention suggestions based on the individual's real-time state characteristics, historical behavioral patterns, and evolving needs. This hinders the effective application of intelligent care technology in the precise health management of dementia patients.

[0004] In view of this, the present invention proposes an adaptive care decision system for elderly people with dementia based on dynamic needs profiles to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an adaptive care decision-making system for dementia-affected elderly based on dynamic demand profiling, comprising: Data acquisition and cleaning module: Collects multi-source vital signs and spatial location signals of elderly people with dementia, performs data cleaning operations on the multi-source vital signs and spatial location signals, and outputs the original dataset of individuals; Multidimensional Feature Extraction Module: This module extracts multidimensional features from the original dataset, separating physiological rhythm features, cognitive decline features, behavioral trajectory features, and emotional fluctuation features, and integrating them into a feature set corresponding to elderly people with dementia. Physiological rhythm features include: heart rate peak-to-trough difference, temporal distribution of heart rate variability, and heart rate baseline offset; cognitive decline features include: variability in daytime visit order and repetition rate of inter-regional transfer paths; behavioral trajectory features include: gait frequency stability, stride symmetry, and gait start-stop response delay; and emotional fluctuation features include: electrical conductance response amplitude and electrical conductance recovery time constant under various activity scenarios. Multi-layer profile construction module: Based on the feature dimensions contained in the human feature set, the profile dimension level is set, and according to the correlation mapping relationship between the profile dimension level and each feature dimension in the human feature set, a multi-layer needs profile of elderly people with dementia is constructed. Dynamic source tracing and iteration module: Real-time acquisition of real-time monitoring data of elderly people with dementia, dynamic state recognition based on real-time monitoring data and multi-layered needs profiles of elderly people with dementia, output profile state features; based on profile state features, causal source analysis of the symptoms corresponding to the multi-layered needs profiles of elderly people with dementia is performed to obtain causal source tracing results. Decision generation and distribution module: Based on the causal tracing results, it selects an appropriate care plan from the care strategy library, performs personalized adjustments on the appropriate care plan to generate adaptive care decisions, and distributes the adaptive care decisions to various receiving terminals for storage and collaborative interaction.

[0006] The technical effects and advantages of the adaptive care decision-making system for dementia elderly based on dynamic demand profiling of this invention are as follows: This invention comprehensively characterizes the state of elderly people with dementia across different dimensions by separating physiological rhythm features, cognitive decline features, behavioral trajectory features, and emotional fluctuation features from original datasets. Given the fundamental differences in baseline levels and change trajectories among individual elderly people with dementia, this invention constructs a multi-layered needs profile based on a set of individual features, including a physiological baseline layer, a cognitive state layer, a behavioral pattern layer, and an emotional representation layer. This establishes individualized physiological baseline intervals and behavioral benchmark maps for elderly people with dementia, overcoming the limitations of traditional fixed-threshold monitoring methods that cannot accurately identify individualized abnormal states. Furthermore, the state of elderly people with dementia exhibits a gradual evolution as the disease progresses. By acquiring real-time monitoring data and performing dynamic state recognition with the multi-layered needs profile, the invention outputs the profile's state features. By combining a knowledge graph of symptom triggers for causal analysis, the system achieves accurate identification and etiological localization of abnormal states, upgrading from passive threshold alarms to proactive intelligent decision-making. Based on the causal analysis results, suitable care plans are selected from the care strategy library, and personalized adjustments are made according to the daily activity patterns and emotionally sensitive periods of elderly people with dementia, generating adaptive care decisions that precisely match care responses with individual needs. These adaptive care decisions are distributed to caregiver terminals, smart device terminals, and remote medical terminals according to the type of implementer. Through a collaborative interactive status table, real-time synchronization of care information and coordination of execution progress are achieved, improving the efficiency of multi-terminal collaborative care and promoting the effective application of intelligent care technology in the precise health management of elderly people with dementia. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the adaptive care decision-making system for elderly people with dementia based on dynamic demand profiling, as described in this invention. Detailed Implementation

[0008] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0009] Example 1 Please see Figure 1 As shown, this embodiment of the adaptive care decision-making system for elderly people with dementia based on dynamic needs profiling includes: Data acquisition and cleaning module: Collects multi-source vital signs and spatial location signals of elderly people with dementia, performs data cleaning operations on the multi-source vital signs and spatial location signals, and outputs the original dataset of individuals.

[0010] Dementia patients exhibit complex symptoms such as memory impairment, disorientation, behavioral abnormalities, and mood swings due to progressive cognitive decline. Their care needs are highly individualized and dynamically changing. To achieve personalized and precise care for elderly people with dementia, it is necessary to comprehensively and continuously acquire multi-dimensional physiological and behavioral data of them as the data foundation for subsequent individualized needs profiling and adaptive care decision-making.

[0011] In this embodiment of the invention, a wearable sensor array deployed on the body surface of an elderly person with dementia continuously collects multi-source vital signs signals, including heart rate variability sequences, body surface temperature sequences, skin conductance sequences, and limb acceleration sequences. Specifically, the heart rate variability sequence is continuously acquired using an ECG patch sensor attached to the chest of the elderly person with dementia at a sampling frequency of 250 times per second, reflecting the activity state of the autonomic nervous system and changes in cardiovascular function; the body surface temperature sequence is acquired using a temperature sensor worn on the wrist of the elderly person with dementia at a sampling frequency of 1 time per minute; the skin conductance sequence is acquired using a conductance sensor attached to the fingertips of the elderly person with dementia at a sampling frequency of 10 times per second; skin conductance reflects the activity intensity of the sympathetic nervous system, and its changes are closely related to the level of emotional arousal, which can be used to indirectly assess the emotional fluctuations of the elderly person with dementia; the limb acceleration sequence is acquired using a triaxial accelerometer fixed to the waist of the elderly person with dementia at a sampling frequency of 50 times per second.

[0012] Simultaneously, spatial location signals are acquired through a personal positioning tag. These signals include the real-time coordinates and movement speed of the elderly person with dementia within the care area. In this embodiment, the personal positioning tag employs ultra-wideband indoor positioning technology. Real-time positioning is achieved through a UWB base station array deployed within the care area, achieving centimeter-level accuracy. The sampling frequency is set to once per second. The real-time coordinates in the spatial location signal represent the two-dimensional position coordinates of the elderly person with dementia within the planar coordinate system of the care area, and the movement speed is the instantaneous speed obtained by dividing the change in coordinates between adjacent sampling moments by the sampling time interval.

[0013] It should be noted that in other embodiments of the present invention, the wearable sensor array may also use wearable devices of different forms such as smart bracelets, smart chest straps or smart insoles to collect vital signs signals; the positioning technology may also use Bluetooth low power beacon positioning or other methods. The implementer may choose according to the specific conditions and accuracy requirements of the care scenario, and no limitation is made here.

[0014] Because wearable sensors are inevitably affected by factors such as sensor detachment, electromagnetic interference, and fluctuations in the positioning system during the daily activities of elderly people with dementia, invalid data is mixed into the raw signals collected. If this invalid data is not effectively cleaned, it will seriously affect the accuracy of subsequent feature extraction and profile construction. Therefore, it is necessary to perform targeted data cleaning operations on multi-source vital sign signals and spatial location signals separately.

[0015] Specifically, the multi-source vital signs signals are separated into channels according to the acquisition channel identifier, that is, the heart rate variability sequence, body surface temperature sequence, skin conductance sequence and limb acceleration sequence are respectively assigned to their corresponding independent data channels; the signal continuity of the vital signs signal in each channel is detected, and the specific detection method is as follows: a sliding window is set on the time series of the vital signs signal in each channel, the length of the sliding window is set to 30 seconds, the sliding step size is set to 5 seconds, and the continuity of the signal sampling points is detected within each sliding window. In this embodiment, signal continuity detection includes the identification of two types of anomalies: The first type is signal interruption segments caused by sensor detachment, characterized by a duration of consecutive missing sampling points within the sliding window exceeding a preset interruption threshold. In this embodiment, the preset interruption threshold is set to 3 seconds. Under normal wearing conditions, the signal collected by the sensor will not have a continuous loss exceeding 3 seconds. If a continuous signal loss exceeds 3 seconds, it can be determined that the signal interruption is caused by sensor detachment or poor contact. The second type is signal distortion segments caused by electromagnetic interference, characterized by an instantaneous change rate of signal amplitude within the sliding window exceeding 5 times the normal physiological change rate of the channel signal. In this embodiment, the normal physiological change rate of each channel is determined based on the 95th percentile of the statistical distribution of the change rate of the effective signal in the previous 24 hours for that channel. The basis for setting this multiple is that the normal fluctuation of physiological signals usually does not exceed 5 times the statistical baseline. An instantaneous change rate exceeding this multiple can be determined as signal distortion caused by electromagnetic interference or a sudden change in sensor contact potential. The identified signal interruption segments and signal distortion segments are marked as invalid signal intervals and discarded.

[0016] It should be noted that in other embodiments of the present invention, the preset interruption threshold can be adjusted within the range of 1 to 10 seconds according to the sensor type and sampling frequency, and the multiple threshold of the normal physiological change rate can be adjusted within the range of 3 to 8 times. The implementer can set it according to the sensor accuracy and the electromagnetic environment of the care scenario.

[0017] Position drift detection is performed on the spatial location signal. Specifically, the coordinate change between adjacent sampling times is calculated, i.e., the Euclidean distance between the position coordinates of two adjacent sampling times. This Euclidean distance is divided by the sampling time interval to obtain the instantaneous movement speed between adjacent sampling times. By comparing this instantaneous movement speed with the physiological movement limit speed of elderly people with dementia, it is determined whether there are abnormal jumps in the positioning data. In this embodiment, the physiological movement limit speed is set to 2 meters per second. The basis for setting this threshold is that, due to age and cognitive function limitations, the movement speed of elderly people with dementia, even when walking quickly or running at a small distance, usually does not exceed 2 meters per second. If the calculated instantaneous movement speed exceeds this limit speed, it indicates that the coordinate change between adjacent sampling times exceeds the physiological displacement range achievable by elderly people with dementia. Coordinate jumps exceeding the physiological movement limit speed are marked as positioning anomalies and removed.

[0018] It should be noted that, in other embodiments of the present invention, the physiological speed limit can be adjusted within the range of 1.5 m / s to 5 m / s according to the specific physical condition and motor ability of the elderly with dementia.

[0019] The multi-source vital signs and spatial location signals, after the above cleaning process, are sequentially arranged according to a unified timestamp, with a minimum time granularity of 1 second. All effective signal data from all channels are aligned to the same time axis to output the original human dataset. Each data record in the original human dataset includes: timestamp, heart rate, body surface temperature, skin conductance, triaxial acceleration, spatial coordinates, and movement speed.

[0020] Multidimensional feature extraction module: Performs multidimensional feature extraction on the original human dataset, separating physiological rhythm features, cognitive decline features, behavioral trajectory features and emotional fluctuation features from the original human dataset, and integrating them into a human feature set corresponding to elderly people with dementia.

[0021] Because the care needs of elderly people with dementia involve multiple dimensions, including physiological, cognitive, behavioral, and emotional aspects, and the characteristics of different dimensions reflect the health status and changes in needs of elderly people with dementia from different perspectives, a single-dimensional feature cannot fully depict the complex state of elderly people with dementia. Therefore, it is necessary to systematically extract multi-dimensional features from the original dataset to lay the foundation for subsequently constructing a multi-layered needs profile.

[0022] Specifically, heart rate variability sequences are extracted from the original human dataset, and segmented statistically analyzed according to the circadian rhythm. In this embodiment, the circadian rhythm is set to 24 hours, with the start time of each circadian cycle set to 00:00:00 daily and the end time set to 23:59:59 daily. The following statistical operations are performed on the heart rate variability sequences within each circadian cycle: (1) Calculate the peak-to-trough difference of heart rate in each 24-hour cycle: In the heart rate variation sequence of a 24-hour cycle, extract the maximum and minimum heart rate values, and record the difference between them as the peak-to-trough difference of heart rate in that 24-hour cycle. The peak-to-trough difference of heart rate reflects the range of heart rate fluctuations in elderly people with dementia in a complete 24-hour cycle. The peak-to-trough difference of heart rate in normal people usually remains within a relatively stable range. If the peak-to-trough difference of heart rate continues to increase or decrease, it may indicate that the autonomic nervous regulation function of elderly people with dementia has changed.

[0023] (2) Calculate the temporal distribution of heart rate variability: Divide a 24-hour cycle into several time segments of equal length. In this embodiment, the length of each time segment is set to 1 hour, so each 24-hour cycle contains 24 time segments. Calculate the standard deviation of the heart rate variation sequence in each time segment to obtain 24 standard deviation values. These 24 standard deviation values ​​are arranged in chronological order to form the temporal distribution of heart rate variability.

[0024] (3) Calculate the baseline heart rate shift between adjacent diurnal cycles: Calculate the mean of the heart rate variation sequence for two adjacent diurnal cycles, and record the absolute value of the difference between the two means as the baseline heart rate shift between adjacent diurnal cycles. The baseline heart rate shift reflects the degree of drift of the baseline heart rate level of elderly people with dementia during the day. If the baseline heart rate shift continues to increase, it may indicate that the basal metabolic level or cardiovascular function of elderly people with dementia is undergoing gradual changes.

[0025] The temporal distribution of heart rate peak-to-trough difference, heart rate variability, and heart rate baseline offset are combined to form physiological rhythm characteristics.

[0026] The movement trajectory sequence corresponding to the spatial location signal is extracted from the original person dataset. The movement trajectory sequence is a time-series coordinate sequence formed by arranging the real-time coordinates of the elderly with dementia in the care area in chronological order. A dwelling area clustering operation is performed on the movement trajectory sequence. Specifically, trajectory segments in the movement trajectory sequence whose movement speed is lower than a preset dwelling speed threshold and whose duration exceeds a preset dwelling time threshold are identified as dwelling events. In this embodiment, the preset dwelling speed threshold is set to 0.1 m / s, and the preset dwelling time threshold is set to 120 seconds. When the movement speed of the elderly with dementia is lower than 0.1 m / s and the duration exceeds 120 seconds, it can be considered that the elderly with dementia are staying at that location rather than experiencing a brief deceleration during walking. The center coordinates of all dwelling events are used as clustering input, and the DBSCAN density clustering algorithm is used to spatially cluster the center coordinates of the dwelling events. The neighborhood radius parameter of the DBSCAN algorithm is set to 2 meters, and the minimum sample number parameter is set to 3. Each cluster obtained after clustering corresponds to a dwelling area within the care area.

[0027] It should be noted that in other embodiments of the present invention, the clustering algorithm may also adopt K-means clustering or hierarchical clustering, the neighborhood radius parameter may be adjusted in the range of 1 meter to 5 meters, and the minimum number of samples parameter may be adjusted in the range of 2 to 5. The implementer may set these parameters according to the spatial layout of the care area.

[0028] The total number of visits by elderly people with dementia to each residence area was counted in each diurnal cycle. Residence areas with more visits than the average number of visits in all diurnal cycles were marked as high-frequency residence areas, and residence areas with fewer visits than the average number of visits were marked as low-frequency residence areas. The cognitive decline of elderly people with dementia is usually manifested as a reduction in activity range, increased dependence on familiar environments, and a decline in the ability to explore new environments. The changes in the distribution of high-frequency residence areas and low-frequency residence areas can indirectly reflect the evolution trend of cognitive function.

[0029] The diurnal variation in the order of visits to high-frequency residence areas is calculated as follows: Within each diurnal cycle, the order in which elderly individuals with dementia visit each high-frequency residence area is recorded, forming a daily visit sequence. The edit distance (Levenshtein distance) between the visit sequence sequences of two adjacent diurnal cycles is calculated. This edit distance is then normalized by dividing it by the average length of the visit sequence; the normalized result is the diurnal variation in the order of visits. The repetition rate of inter-regional transfer paths is calculated as follows: Transfer records of elderly individuals with dementia between adjacent residence areas are extracted within each diurnal cycle, forming a set of regional transfer paths. The proportion of transfer paths in the current diurnal cycle that are identical to those in the previous diurnal cycle's set is calculated out of the total number of transfer paths in the current diurnal cycle; this proportion is the repetition rate of inter-regional transfer paths. The repetition rate reflects the stability of the behavioral patterns of elderly individuals with dementia: A continuously decreasing repetition rate suggests that the spatial navigation ability and regularity of daily activities of elderly individuals with dementia are weakening, potentially indicating further cognitive decline. The diurnal variation in the order of visits and the repetition rate of inter-regional transfer paths are combined to form a characteristic of cognitive decline.

[0030] Limb acceleration sequences were extracted from the original human dataset. These sequences were time-series accelerations acquired by a triaxial accelerometer along the forward / backward (X-axis), left / right (Y-axis), and vertical (Z-axis) directions. Gait cycle segmentation was performed on the limb acceleration sequences. Specifically, bandpass filtering was applied to the vertical acceleration sequences. In this embodiment, the passband frequency range of the bandpass filter was set to 0.5Hz to 3Hz. Components below 0.5Hz were low-frequency noise such as gravity shifts, while components above 3Hz were high-frequency noise such as sensor vibrations. Peak detection was then performed on the filtered vertical acceleration sequences, and the time interval between two adjacent peaks was taken as a gait cycle, thus segmenting the limb acceleration sequences into continuous gait cycle segments.

[0031] It should be noted that, in other embodiments of the present invention, the passband frequency range of the bandpass filter can be adjusted within the range of 0.3Hz to 5Hz according to the walking speed characteristics of elderly people with dementia, and the gait period segmentation can also be performed using methods such as zero-crossing detection or autocorrelation function, which are not limited here.

[0032] Perform the following feature extraction for each gait cycle: (1) Extracting gait frequency stability: Calculate the coefficient of variation (i.e., standard deviation divided by mean) of the cycle duration sequence of multiple consecutive gait cycles. The smaller the coefficient of variation, the more stable the gait frequency. In this embodiment, the coefficient of variation is calculated using 20 consecutive gait cycles as an analysis window. 20 gait cycles correspond to a walking process of about 10 to 15 seconds, which is sufficient to reflect the gait frequency stability during a continuous walking process.

[0033] (2) Extract stride symmetry: Integrate the forward and backward acceleration sequence in each gait cycle to obtain the stride estimate; divide the absolute value of the difference between the stride estimates of two adjacent gait cycles by the average of the two to obtain the stride symmetry index. The closer the index is to zero, the more symmetrical the stride.

[0034] (3) Extracting gait start-stop response delay: Identify the moments when elderly people with dementia start walking from a stationary state and stop walking from a stationary state. Calculate the time difference from the start command (marked by the beginning of regular fluctuations in the acceleration signal) to reaching a stable gait, and the time difference from the stop command (marked by the beginning of decay of the acceleration signal fluctuations) to complete stillness. The average of the start delay and the stop delay is taken as the gait start-stop response delay. The gait start-stop response delay reflects the motor response ability and motor control ability of elderly people with dementia. Combine gait frequency stability, stride symmetry, and gait start-stop response delay into behavioral trajectory features.

[0035] Skin conductance sequences were extracted from the original dataset of individuals. Changes in skin conductance are closely related to the activity of the sympathetic nervous system. When an individual is in a state of emotional arousal, such as tension, anxiety, or excitement, the secretion of sweat glands increases, leading to an increase in skin conductance. When emotions return to calm, skin conductance gradually returns to baseline levels. Therefore, skin conductance signals can serve as an indirect indicator for assessing the emotional fluctuations of elderly people with dementia.

[0036] The conductivity response amplitude and conductivity recovery time constant of the skin conductivity sequence under different activity scenarios were calculated. First, the activity scenarios of the elderly with dementia at different time periods were determined based on the spatial location signals in the original population dataset. In this embodiment, the activity scenarios include dining scenarios, resting scenarios, activity scenarios, and social scenarios. The determination of the activity scenario is based on the functional area category corresponding to the location coordinates of the elderly with dementia.

[0037] For each activity scenario, perform the following feature calculations: (1) Calculation of conductance response amplitude: Within 10 minutes after the elderly person with dementia enters the activity scene, the difference between the maximum value of the skin conductance sequence and the mean value of the skin conductance sequence within 1 minute before entering the scene is extracted. This difference is the conductance response amplitude under the activity scene. The conductance response amplitude reflects the intensity of the elderly person with dementia's emotional response to a specific activity scene.

[0038] (2) Calculation of the conductance recovery time constant: After the elderly person with dementia leaves the activity scene, the time required for the skin conductance sequence to fall from the peak to the baseline level (the baseline level is the average of the skin conductance sequence within 1 minute before entering the scene) is extracted. This time is the conductance recovery time constant for that activity scene. The conductance recovery time constant reflects the strength of the elderly person's emotional regulation ability: the longer the conductance recovery time constant, the longer it takes for the elderly person with dementia to recover from an emotional arousal state to a calm state, and the weaker their emotional regulation ability. The conductance response amplitude and conductance recovery time constant under each activity scene are combined to form the emotional fluctuation characteristics.

[0039] Integration of Character Feature Sets: Physiological rhythm features, cognitive decline features, behavioral trajectory features, and emotional fluctuation features are structurally encapsulated according to feature dimension identifiers and integrated into a character feature set. The specific method of structured encapsulation is as follows: a unique dimension identifier code is assigned to each feature dimension. The dimension identifier code for physiological rhythm features is "PHY", the dimension identifier code for cognitive decline features is "COG", the dimension identifier code for behavioral trajectory features is "BEH", and the dimension identifier code for emotional fluctuation features is "EMO". Each dimension identifier code is associated with the specific feature data corresponding to that dimension, forming a hierarchical character feature set.

[0040] By systematically extracting multi-dimensional features, the raw monitoring data of elderly people with dementia is transformed into structured feature representations with clear semantics, which solves the technical limitation of traditional monitoring systems that rely on only a single-dimensional indicator for anomaly judgment.

[0041] Multi-layer profile construction module: Based on the feature dimensions contained in the human feature set, the profile dimension level is set, and according to the correlation mapping relationship between the profile dimension level and each feature dimension in the human feature set, a multi-layer needs profile of elderly people with dementia is constructed.

[0042] Existing dementia care monitoring systems typically use standardized parameters to identify abnormalities in all elderly individuals, lacking the ability to establish dynamic baselines tailored to individual differences. This leads to a systematic discrepancy between care responses and individual needs. This step addresses this by establishing individualized baseline parameters across different feature dimensions, constructing a multi-layered needs profile that dynamically reflects the individual state of dementia patients. This provides an individualized reference standard for subsequent personalized anomaly identification and care decisions.

[0043] The profile hierarchy is structured as follows: Corresponding profile hierarchy levels are assigned to the physiological rhythm characteristics, cognitive decline characteristics, behavioral trajectory characteristics, and emotional fluctuation characteristics included in the individual's feature set. The profile hierarchy comprises a physiological baseline layer, a cognitive state layer, a behavioral pattern layer, and an emotional representation layer. The physiological baseline layer corresponds to physiological rhythm characteristics, the cognitive state layer to cognitive decline characteristics, the behavioral pattern layer to behavioral trajectory characteristics, and the emotional representation layer to emotional fluctuation characteristics. These four layers depict the individualized state of elderly people with dementia from different perspectives, and are interconnected and complementary: the physiological baseline layer reflects the basic physiological state of elderly people with dementia, the cognitive state layer reflects the cognitive function stage of elderly people with dementia, the behavioral pattern layer reflects the daily activity patterns of elderly people with dementia, and the emotional representation layer reflects the emotional response characteristics of elderly people with dementia.

[0044] Construction of the physiological baseline layer: In the physiological baseline layer, the heart rate baseline offset in the physiological rhythm characteristics is used as the anchoring parameter to establish an individualized physiological baseline interval for elderly people with dementia. Specifically, heart rate peak-to-trough difference data from elderly people with dementia are collected over multiple consecutive 24-hour cycles. In this embodiment, seven consecutive 24-hour cycles are used as the initial observation window. These seven cycles constitute a complete weekly calendar, covering different activity patterns of elderly people with dementia on weekdays and rest days, ensuring sufficient representativeness of the statistical results. The mean of the heart rate peak-to-trough difference data from the seven 24-hour cycles is calculated. with standard deviation ,use As the upper and lower bounds of the individualized physiological baseline interval.

[0045] Construction of the Cognitive State Layer: In the cognitive state layer, the degree of change in daytime visit order and the repetition rate of inter-regional transfer paths, which are characteristics of cognitive decline, are used as evaluation indicators to label the current cognitive function stage of elderly people with dementia. The specific labeling method is as follows: Based on the clinical dementia staging criteria, the cognitive function stage is divided into three stages: mild cognitive impairment, moderate cognitive impairment, and severe cognitive impairment. Stage classification thresholds are set for the degree of change in daytime visit order and the repetition rate of inter-regional transfer paths: when the degree of change in daytime visit order is less than 0.3 and the repetition rate of inter-regional transfer paths is greater than 0.7, it is labeled as mild cognitive impairment; when the degree of change in daytime visit order is between 0.3 and 0.6 and the repetition rate of inter-regional transfer paths is between 0.4 and 0.7, it is labeled as moderate cognitive impairment; when the degree of change in daytime visit order is greater than 0.6 and the repetition rate of inter-regional transfer paths is less than 0.4, it is labeled as severe cognitive impairment. It should be noted that the above threshold is an empirical setting value for this embodiment. In other embodiments of the present invention, it can be adjusted according to the clinical assessment standards of care institutions and the specific course of the disease of elderly people with dementia.

[0046] Construction of the Behavioral Pattern Layer: In the behavioral pattern layer, gait frequency stability and stride symmetry, which are characteristics of behavioral trajectories, are used as parameters to establish an individualized behavioral baseline map for elderly people with dementia. The specific construction method is as follows: Gait frequency stability and stride symmetry data of elderly people with dementia are collected over seven consecutive 24-hour cycles. Each hour of the 24-hour period is considered a time segment, and the mean values ​​of gait frequency stability and stride symmetry are calculated for each time segment. The sequence of mean gait frequency stability and mean stride symmetry values ​​corresponding to the 24 time segments are arranged chronologically to form a temporal pattern map of the daily activities of elderly people with dementia, i.e., the individualized behavioral baseline map.

[0047] Construction of the Emotion Representation Layer: In the emotion representation layer, the conductivity response amplitude and conductivity recovery time constant, which are key characteristics of emotional fluctuations, are used as parameters to establish a ground state distribution map of emotions in elderly individuals with dementia. Specifically, the construction method involves collecting data on conductivity response amplitude and conductivity recovery time constant under various activity scenarios for seven consecutive 24-hour cycles. The mean and standard deviation of the conductivity response amplitude and the mean and standard deviation of the conductivity recovery time constant are calculated for each activity scenario. The mean conductivity response amplitude and mean conductivity recovery time constant for each activity scenario are then arranged according to the activity scenario category to form the ground state distribution map of emotions in elderly individuals with dementia. The emotional ground state distribution map includes markers for emotionally sensitive periods. The criteria for determining emotionally sensitive periods are as follows: if the average conductance response amplitude within a certain time segment exceeds 1.5 times the total average conductance response amplitude across all time segments, or the average conductance recovery time constant exceeds 1.5 times the total average conductance recovery time constant across all time segments, then that time segment is marked as an emotionally sensitive period. An emotional reaction intensity or recovery time exceeding 1.5 times the total average can be considered as indicating that the elderly person with dementia is significantly more sensitive to emotional stimuli than their daily baseline level during that period. It should be noted that in other embodiments of this invention, the multiple for determining emotionally sensitive periods can be adjusted within the range of 1.2 to 2.0.

[0048] Assembly of Multi-Layer Needs Profiles: Individualized physiological baseline intervals, cognitive function stage labels, individualized behavioral baseline maps, and emotional baseline distribution maps are assembled in a hierarchical nested structure to construct multi-layered needs profiles for elderly individuals with dementia. The assembly method for the hierarchical nested structure is as follows: The unique identifier of the elderly person with dementia serves as the root node of the profile. Under the root node, profile data for each level is sequentially attached according to the priority of the profile dimensions (physiological baseline layer -> cognitive state layer -> behavioral pattern layer -> emotional representation layer), forming a hierarchical structure. A temporal relationship is established between each level through a unified time label, enabling joint querying and analysis of data from different levels within the profile for the same period.

[0049] This embodiment establishes individualized baseline parameters across multiple dimensions, constructing a multi-layered needs profile that comprehensively depicts the individual state of elderly people with dementia. This solves the problem that traditional systems using uniform standardized parameters cannot adapt to individual differences, providing an individualized reference benchmark for subsequent personalized dynamic state recognition and care decisions.

[0050] Dynamic source tracing and iteration module: Real-time acquisition of real-time monitoring data of elderly people with dementia, dynamic state recognition based on real-time monitoring data and multi-layered needs profiles of elderly people with dementia, outputting profile state features; based on the profile state features, causal source analysis of the symptoms and causes corresponding to the multi-layered needs profiles of elderly people with dementia is performed to obtain causal source tracing results.

[0051] After constructing a multi-layered needs profile of elderly people with dementia, it is necessary to continuously monitor their real-time status changes. By comparing the real-time monitoring data with the individualized baseline in the multi-layered needs profile, abnormal states that deviate from the individual baseline can be identified. Further, the underlying causes of these abnormal states can be traced and analyzed, providing a basis for subsequent precise care decisions. This dynamic state identification method based on individualized baselines, compared to traditional fixed threshold alarm methods, can effectively reduce false alarm and false negative rates and improve the accuracy of abnormal state identification.

[0052] Specifically, real-time monitoring data is acquired through a wearable sensor array carried by elderly people with dementia. The real-time monitoring data includes real-time heart rate, real-time body surface temperature, real-time skin conductivity, real-time limb acceleration, and real-time spatial coordinates within the current acquisition period. After the real-time monitoring data is acquired, the same data cleaning operations as mentioned above are required, including signal continuity detection, electromagnetic interference signal distortion removal, and positioning drift detection, to ensure the validity of the real-time monitoring data.

[0053] Perform dynamic state identification: Dynamic state identification involves comparing real-time monitoring data layer by layer with the individualized baselines at each level of the multi-layered needs profile of elderly people with dementia, identifying abnormal indicators that deviate from the individualized baselines. The specific process is as follows: The system assigns real-time heart rate values ​​to the individualized physiological baseline intervals recorded in the multi-layered needs profile of elderly people with dementia. Specifically, it calculates the mean of the real-time heart rate sequence within the current collection period and determines whether this mean falls within the individualized physiological baseline interval. If it does, no physiological deviation indicator is generated. When the mean of the real-time heart rate falls outside the individualized physiological baseline interval, the deviation between the mean of the real-time heart rate and the nearest boundary of the individualized physiological baseline interval is calculated. The deviation is calculated as follows: if the mean of the real-time heart rate is greater than the upper boundary of the individualized physiological baseline interval, the deviation is equal to the difference between the mean of the real-time heart rate and the upper boundary; if the mean of the real-time heart rate is less than the lower boundary of the individualized physiological baseline interval, the deviation is equal to the difference between the lower boundary and the mean of the real-time heart rate. Dividing the deviation by the width of the individualized physiological baseline interval gives the magnitude of the physiological deviation indicator. Simultaneously record the direction of deviation: if the mean instantaneous heart rate is higher than the upper limit, the deviation direction is marked as positive; if it is lower than the lower limit, the deviation direction is marked as negative. The magnitude and direction of deviation together constitute the physiological deviation indicator.

[0054] The system performs trajectory matching calculations between real-time spatial coordinate values ​​and the individualized behavioral baseline map recorded in the behavioral pattern layer of the multi-layered needs profile of elderly people with dementia. Specifically, based on the time segment corresponding to the current collection period (determined by hour), the system extracts the historical regular path corresponding to that time segment from the individualized behavioral baseline map. The historical regular path represents the typical activity trajectory of the elderly person with dementia within that time segment. The system then extracts the current movement path composed of real-time spatial coordinate values ​​within the current collection period and calculates the path deviation between the current movement path and the historical regular path. The path deviation is calculated using a dynamic time warping algorithm to calculate the DTW distance between the current movement path and the historical regular path. Dividing the DTW distance by the total length of the historical regular path gives the path deviation. When the path deviation exceeds a preset behavioral deviation threshold, a behavioral deviation indicator is generated. In this embodiment, the preset behavioral deviation threshold is set to 0.5. This threshold is set based on the fact that a path deviation exceeding 0.5 indicates a significant difference between the current movement path and the historical regular path, suggesting a significant deviation in the behavioral pattern of the elderly person with dementia. It should be noted that in other embodiments of the present invention, the trajectory matching algorithm may also employ methods such as Hausdorff distance, and the preset behavior deviation threshold may be adjusted within the range of 0.3 to 0.8.

[0055] The instantaneous skin conductance values ​​are compared with the emotional base state distribution map recorded in the emotional representation layer of the multi-layered needs profile of elderly people with dementia. Specifically, based on the activity scenario of the elderly person with dementia during the current collection period, the base state conductance range corresponding to that activity scenario is extracted from the emotional base state distribution map. The base state conductance range is the interval consisting of the mean of the conductance response amplitude in that activity scenario plus or minus one standard deviation. The deviation between the mean of the instantaneous skin conductance values ​​and the base state conductance range during the current collection period is calculated: if the mean of the instantaneous skin conductance values ​​falls within the base state conductance range, no emotional deviation indicator is generated; if it falls outside the base state conductance range, the deviation is calculated as the absolute value of the difference between the mean of the instantaneous skin conductance values ​​and the nearest boundary of the base state conductance range divided by the width of the base state conductance range. When the deviation exceeds a preset emotional deviation threshold, an emotional deviation indicator is generated. In this embodiment, the preset emotional deviation threshold is set to 0.8. A deviation exceeding 0.8 indicates a significant deviation between the current emotional response and the base state level, requiring attention from caregivers. It should be noted that, in other embodiments of the present invention, the preset emotion offset threshold can be adjusted within the range of 0.5 to 1.5.

[0056] Physiological deviation indicators, behavioral deviation indicators, and emotional deviation indicators are arranged in a structured manner according to indicator categories. Each indicator includes an indicator category identifier, deviation magnitude value, deviation direction marker, and detection timestamp, outputting the profile status features.

[0057] After obtaining the characteristics of the dementia patient's condition, it is necessary to conduct a source analysis on the abnormal conditions reflected in the characteristics of the dementia patient's condition to determine the possible causes of the abnormal conditions, so as to provide accurate etiological basis for subsequent care decisions.

[0058] Specifically, physiological deviation indicators, behavioral deviation indicators, and emotional deviation indicators are extracted from the portrait status features, and a correlation analysis is performed on the temporal co-occurrence relationship among these three indicators. The method for analyzing the temporal co-occurrence relationship is as follows: a temporal co-occurrence window is set; in this embodiment, the length of the temporal co-occurrence window is set to 30 minutes. Within the temporal co-occurrence window, combinations of indicators that simultaneously deviate are identified: if two or more of the physiological deviation indicators, behavioral deviation indicators, and emotional deviation indicators appear simultaneously within the same 30-minute temporal co-occurrence window, these simultaneously appearing indicator combinations are marked as a linked deviation pattern. The linked deviation pattern records which indicator categories deviate simultaneously, the direction of deviation for each indicator, and the magnitude of deviation for each indicator.

[0059] The linkage deviation pattern is matched with a pre-built knowledge graph of dementia symptom triggers. This knowledge graph is a pre-constructed domain knowledge base that records the mapping relationships between various known symptom triggers and their corresponding multi-dimensional indicator linkage features. In this embodiment, the knowledge graph includes, but is not limited to, the following symptom trigger entries: anxiety attacks (corresponding standard indicator linkage features: positive deviation of physiological deviation indicators + positive deviation of emotional deviation indicators), hypoglycemic events (corresponding standard indicator linkage features: positive deviation of physiological deviation indicators + positive deviation of behavioral deviation indicators), daytime sleepiness caused by sleep disorders (corresponding standard indicator linkage features: negative deviation of physiological deviation indicators + negative deviation of behavioral deviation indicators), pain and discomfort (corresponding standard indicator linkage features: positive deviation of physiological deviation indicators + positive deviation of emotional deviation indicators + negative deviation of behavioral deviation indicators), etc.

[0060] It should be noted that the content of the knowledge graph on the causes of dementia symptoms was compiled by medical experts based on clinical experience and medical literature. In other embodiments of the present invention, it can be expanded and adjusted according to the clinical practice experience of care institutions.

[0061] The specific process of graph matching is as follows: From the linked deviation pattern, extract the combination of indicator categories that simultaneously deviate and the deviation direction markers for each indicator. The deviation direction markers are distinguished as positive and negative deviations. Combine the indicator category combinations and the deviation direction markers for each indicator into a linked feature vector. Read each symptom trigger entry one by one from the dementia symptom trigger knowledge graph. Each symptom trigger entry contains the standard indicator linked features corresponding to that symptom trigger. The standard indicator linked features record the indicator category and corresponding deviation direction that should occur when the symptom trigger is triggered. Perform an item-by-item comparison operation between the linked feature vector and the standard indicator linked features of each symptom trigger entry. The item-by-item comparison operation includes two levels of comparison: The first level is to determine whether the indicator categories in the linked feature vector are covered by the indicator categories of the standard indicator linked features. That is, whether the indicator categories that deviate in the linked feature vector belong to a subset or an equal set of indicator categories that should deviate as specified in the standard indicator linked features. The indicator category coverage rate is calculated by dividing the number of indicator categories in the linked feature vector covered by the standard indicator linked features by the total number of indicator categories specified in the standard indicator linked features. The second level is to determine whether the deviation direction markers of each indicator in the linked feature vector are consistent with the deviation direction of the corresponding indicator in the standard indicator linked features. The deviation direction consistency rate is calculated by dividing the number of indicators with consistent deviation directions by the total number of indicators in the linked feature vector.

[0062] Items whose symptom category coverage and deviation direction consistency rate both meet preset matching thresholds are identified as candidate symptom causes. In this embodiment, the preset matching threshold for symptom category coverage is set to 0.7, and the preset matching threshold for deviation direction consistency rate is set to 0.8. A coverage rate exceeding 0.7 indicates that more than 70% of the symptom categories in the linkage deviation pattern are included by the standard symptom linkage features, indicating high matching confidence. The deviation direction consistency rate threshold is set based on the following: a consistency rate exceeding 0.8 indicates that more than 80% of the symptom deviation directions are consistent with the standard linkage features, excluding the possibility of mismatches with opposite directions.

[0063] It should be noted that, in other embodiments of the present invention, the preset matching threshold for indicator category coverage can be adjusted in the range of 0.5 to 0.9, and the preset matching threshold for deviation direction consistency rate can be adjusted in the range of 0.6 to 1.0.

[0064] Stage-specific adaptability verification was performed on candidate symptom triggers: Based on the cognitive function stage labels recorded in the cognitive state layer of the multi-layered needs profile of elderly people with dementia, it was determined whether the candidate symptom triggers were consistent with the typical pathological manifestations of the current cognitive function stage of the elderly person with dementia. The specific verification method was as follows: For each symptom trigger entry in the dementia symptom trigger knowledge graph, the probability level of occurrence of the symptom trigger at different cognitive function stages was pre-labeled, with probability levels divided into high, medium, and low. If the probability level of occurrence corresponding to the current cognitive function stage label of the candidate symptom trigger was "high" or "medium," then the adaptability verification was passed; if the probability level was "low," then the candidate symptom trigger was determined to be inconsistent with the current cognitive function stage and failed the adaptability verification. Candidate symptom triggers that passed the adaptability verification were identified as the causal source tracing results.

[0065] By comparing real-time monitoring data with individualized multi-layered needs profiles layer by layer, dynamic anomaly identification based on individualized baselines is achieved, effectively solving the problems of high false alarm and high false negative rates caused by ignoring individual differences in traditional fixed threshold methods. At the same time, through the linkage analysis of multi-dimensional indicators and knowledge graph matching, the source of the disease can be traced from the surface abnormality to the underlying cause of the disease, providing etiological basis for subsequent precision care decisions.

[0066] Decision generation and distribution module: Based on the causal tracing results, it selects an appropriate care plan from the care strategy library, performs personalized adjustments on the appropriate care plan to generate adaptive care decisions, and distributes the adaptive care decisions to various receiving terminals for storage and collaborative interaction.

[0067] Existing dementia care systems can only provide simple, passive threshold alarms, failing to proactively generate targeted care intervention suggestions based on an individual's real-time status characteristics, historical behavioral patterns, and evolving needs. This step, based on causal analysis, precisely selects suitable care plans from a care strategy library and personalizes these plans by considering the individualized behavioral habits and emotional characteristics of the elderly with dementia. Finally, adaptive care decisions are distributed to various terminals, including caregivers, smart devices, and remote healthcare, for collaborative execution, achieving a technological leap from passive alarms to proactive decision-making and from standardized plans to personalized adjustments.

[0068] Specifically, the identified symptom trigger categories and corresponding severity levels of indicators are extracted from the causal tracing results. The severity level is determined by classifying the indicators based on the proportional relationship between the deviation amplitude of each indicator in the profile status characteristics and the corresponding baseline range in the multi-layered needs profile of the dementia patient. Specifically, for physiological deviation indicators, the deviation amplitude is divided by the width of the individualized physiological baseline interval to obtain the amplitude deviation ratio; for behavioral deviation indicators, the path deviation degree is directly used as the behavioral deviation ratio; for emotional deviation indicators, the degree of deviation is used as the emotional deviation ratio. The deviation ratios of all indicators involved in the linked deviations are averaged, and the severity levels are classified according to the mean: a mean less than 1.0 indicates a mild level, a mean between 1.0 and 2.0 indicates a moderate level, and a mean greater than 2.0 indicates a severe level. The criteria for setting these grading thresholds are as follows: a deviation ratio of less than 1.0 indicates that the deviation is less than one time the width of the baseline range, which is considered a slight deviation; a deviation ratio between 1.0 and 2.0 indicates that the deviation is between one and two times the width of the baseline range, which is considered a moderate deviation; and a deviation ratio greater than 2.0 indicates that the deviation is more than twice the width of the baseline range, which is considered a severe deviation.

[0069] It should be noted that, in other embodiments of the present invention, the threshold for classifying the severity of indications can be adjusted based on the clinical experience of medical experts, and the classification can be further refined into five or seven levels, which is not limited here.

[0070] The system employs a primary index based on symptom trigger categories and a secondary index based on severity levels, performing dual retrieval operations within the hierarchical index structure of the care strategy database. The care strategy database is a pre-built database of care plans. It is divided into primary partitions based on symptom trigger categories, such as anxiety attacks, delirium, hypoglycemia events, sleep disorders, and pain / discomfort. Each primary partition is further divided into secondary partitions based on severity levels, specifically mild, moderate, and severe sub-partitions for each symptom trigger category. Each secondary partition stores care plan entries for the corresponding conditions. Each care plan entry includes a care plan number, a sequence of care intervention actions, a care execution time parameter, an applicable cognitive stage range field, and plan effectiveness evaluation indicators.

[0071] It should be noted that the content of the care strategy library is compiled and maintained by experts in the care field based on clinical nursing guidelines and care practice experience, and can be dynamically expanded according to the actual needs of care institutions in other embodiments of the present invention.

[0072] The execution process of the dual search operation is as follows: First, the corresponding primary partition in the care strategy library is located based on the symptom cause category identifier. Then, within the primary partition, the corresponding secondary partition is located based on the severity level of the indication. All care plan entries are retrieved from the secondary partition to form a set of care plan entries.

[0073] From the set of care plan entries returned by the dual search operation, the applicable cognitive stage range field is extracted for each care plan entry. This field records the range of cognitive function stages to which the care plan entry is applicable; for example, the applicable cognitive stage range for a care plan entry might be "mild to moderate." The applicable cognitive stage range field is then matched against the cognitive function stage labels recorded in the cognitive state layer of the multi-layered needs profile for elderly people with dementia. Specifically, the matching method is to determine whether the cognitive function stage label belongs to the set of stages included in the applicable cognitive stage range. If it does, the match is successful; otherwise, the match is unsuccessful. Care plan entries whose applicable cognitive stage range is incompatible with the cognitive function stage label are filtered out, i.e., those that fail to match. The remaining care plan entries are determined as suitable care plans. If multiple care plan entries remain after filtering, the one with the best effectiveness evaluation index is selected as the final suitable care plan.

[0074] The care intervention sequence and execution time parameter are extracted from the adapted care plan. The care intervention sequence contains multiple care intervention action nodes arranged in order of execution. Each care intervention action node corresponds to a specific care intervention operation, such as "playing soothing music" or "guiding the elderly to the rest area." The care execution time parameter includes the suggested execution time and suggested duration for each care intervention action node. The suggested execution time is the time when the care intervention action node is suggested to begin execution, and the suggested duration is the length of time the care intervention action node is suggested to continue execution.

[0075] Personalized adjustments include two aspects: adjusting for time-related conflicts and adjusting the intensity of interventions. Time-based conflict adjustment: Daily activity time-series patterns of elderly people with dementia are extracted from the individualized behavioral baseline map of the behavioral pattern layer in the multi-layered needs profile of elderly people with dementia. These patterns record the typical activity types and duration intervals of elderly people with dementia at different times. Time-based conflict detection is performed between the suggested execution time of each care intervention action node in the care execution time parameter and the daily activity time-series patterns. The specific method for time-based conflict detection is to determine whether the suggested execution time of each care intervention action node falls within the high-activity-density period of the elderly person with dementia. The criteria for determining a high-activity-density period are: in the individualized behavioral baseline map, if the mean gait stability within a certain time segment is less than 0.8 times the total mean gait stability across all time segments (indicating that the elderly person with dementia is active but their gait is not stable during this period), then this time segment is marked as a high-activity-density period. A mean gait stability below 0.8 times the total mean indicates that walking activity is relatively frequent and unstable during this period. Introducing additional care interventions during this period may disrupt the normal activity rhythm of the elderly person with dementia and increase the safety risks such as falls. It should be noted that, in other embodiments of the present invention, the determination factor for high activity density periods can be adjusted within the range of 0.6 to 0.9.

[0076] When the suggested execution time for a care intervention action falls within a high-activity-density period for an elderly person with dementia, the execution time of that care intervention action action is moved to a low-activity-density period adjacent to the high-activity-density period. The low-activity-density period is the nearest non-high-activity-density period before or after the high-activity-density period. The move prioritizes the low-activity-density period following the high-activity-density period. If the subsequent low-activity-density period is too far away (more than 2 hours), the previous low-activity-density period is selected.

[0077] Intervention Intensity Moderation Adjustment: Emotionally sensitive time periods for elderly individuals with dementia are extracted from the emotional base state distribution map of the emotional representation layer in the multi-layered needs profile of dementia patients. Intervention intensity is moderated for care intervention action nodes falling within these emotionally sensitive time periods. Specifically, the adjustment method involves reducing the execution magnitude of environmental change-related operations within the care intervention action nodes. Environmental change-related operations refer to operations that alter the surrounding environment of the elderly person with dementia, including but not limited to adjusting light brightness, sound volume, and spatial layout. The reduction is achieved by multiplying the execution magnitude of environmental change-related operations by a moderation coefficient. In this embodiment, the moderation coefficient is set to 0.6, reducing the original environmental change magnitude to 60%. Reducing the execution magnitude to 60% achieves the purpose of care intervention while avoiding excessive stimulation for the elderly person with dementia.

[0078] It should be noted that in other embodiments of the present invention, the mitigation coefficient can be adjusted in the range of 0.4 to 0.8, and the implementer can set it according to the emotional sensitivity of the elderly with dementia.

[0079] The care intervention sequence, adjusted for time-slot conflicts and intervention intensity mitigation, is repackaged with its corresponding care execution time parameters to generate adaptive care decisions. Each care intervention node in the adaptive care decision includes: action node number, action content description, adjusted execution time, adjusted duration, execution range (adjusted range if environmental change is involved), and execution subject category identifier.

[0080] Terminal adaptability analysis is performed on adaptive care decisions to extract the execution subject category identifiers for each care intervention action node in the adaptive care decision process. The execution subject category identifiers are distinguished into caregiver execution, smart device execution, and remote medical care execution: Caregiver execution corresponds to operations that need to be directly performed by on-site caregivers, such as "guiding the elderly to the rest area"; smart device execution corresponds to operations that can be automatically performed by smart devices within the care area, such as "dimming indoor lights" and "playing soothing music"; remote medical care execution corresponds to operations that require notifying remote medical personnel for assessment or intervention, such as "suggesting remote ECG monitoring".

[0081] Based on the category identifier of the executing entity, adaptive care decisions are broken down into caregiver instruction packages for caregiver terminals, device control instruction packages for smart device terminals, and healthcare notification instruction packages for remote healthcare terminals. The breakdown method is as follows: care intervention action nodes categorized as caregiver execution are grouped into caregiver instruction packages in chronological order of execution; care intervention action nodes categorized as smart device execution are grouped into device control instruction packages in chronological order of execution; and care intervention action nodes categorized as remote healthcare execution are grouped into healthcare notification instruction packages in chronological order of execution.

[0082] A terminal routing identifier and instruction priority label are attached to the caregiver instruction package, device control instruction package, and medical notification instruction package, respectively. The terminal routing identifier points to the communication address of the corresponding receiving terminal: the terminal routing identifier of the caregiver instruction package points to the communication address of the mobile terminal (such as a smartphone or smartwatch) of the currently on-duty caregiver; the terminal routing identifier of the device control instruction package points to the network address of the corresponding smart device in the care area; and the terminal routing identifier of the medical notification instruction package points to the communication address of the remote medical care platform.

[0083] The instruction priority label is determined based on the severity level of the indications at each care intervention node in adaptive care decision-making: if the severity level of the indication is severe, the instruction priority label is set to urgent; if the severity level of the indication is moderate, the instruction priority label is set to important; if the severity level of the indication is mild, the instruction priority label is set to routine.

[0084] The care information distribution channel sequentially pushes caregiver instruction packages, device control instruction packages, and medical notification instruction packages to their respective receiving terminals according to their priority tags. The priority order is as follows: instruction packages tagged "urgent" are pushed first, followed by those tagged "important," and those tagged "regular" are pushed last. Upon receiving the corresponding instruction package, each receiving terminal performs a local storage operation: the caregiver terminal stores the instruction package in its local task list and sends a push notification to the caregiver; the smart device terminal stores the device control instruction package in its local instruction queue and executes it automatically according to the instruction order; the remote medical terminal stores the medical notification instruction package in the to-do list of the remote medical platform and notifies the relevant medical personnel. After completing the storage operation, each receiving terminal sends a reception confirmation signal back to the care information distribution channel. This confirmation signal includes the terminal identifier, instruction package number, and reception timestamp.

[0085] A collaborative interaction status table is established based on the reception confirmation signals returned by each receiving terminal. The collaborative interaction status table uses the decision number of the adaptive care decision as the primary key and records the reception status, execution progress, and completion flag of the corresponding instruction packet for each receiving terminal: the reception status includes three states: "sent," "received," and "reception failed"; the execution progress includes three states: "not started," "in execution," and "completed"; the completion flag is a Boolean value indicating whether all care intervention action nodes in the corresponding instruction packet have been completed. When the execution progress of any receiving terminal changes (e.g., a caregiver begins performing a care task, or a smart device completes a lighting adjustment operation), that receiving terminal sends a progress change signal back to the care information distribution channel. The care information distribution channel updates the execution progress field of the corresponding entry in the collaborative interaction status table and synchronously pushes a progress change notification to other relevant receiving terminals through the collaborative interaction status table. The progress change notification allows each receiving terminal to understand the execution status of other terminals in real time, achieving collaborative cooperation among multiple terminals: for example, when a smart device terminal completes the operation of dimming indoor lights, the caregiver terminal can receive a timely notification, understanding that the environment has been adjusted, thus better executing subsequent operations to guide the elderly to the rest area.

[0086] By selecting precise solutions based on the results of causal tracing, making personalized adjustments based on individual behavioral habits and emotional characteristics, and coordinating distribution and interaction across multiple terminals, a complete closed loop from anomaly identification to precise intervention has been achieved. This solves the technical limitation of existing systems that can only passively alarm and cannot actively generate personalized care decisions, effectively improving the targeting and efficiency of care for elderly people with dementia.

[0087] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive care decision-making system for elderly people with dementia based on dynamic needs profiling, characterized in that, include: Data acquisition and cleaning module: Collects multi-source vital signs and spatial location signals of elderly people with dementia, performs data cleaning operations on the multi-source vital signs and spatial location signals, and outputs the original dataset of individuals; Multidimensional feature extraction module: Performs multidimensional feature extraction on the original human dataset, separating physiological rhythm features, cognitive decline features, behavioral trajectory features and emotional fluctuation features from the original human dataset, and integrating them into a human feature set corresponding to elderly people with dementia; Multi-layer profile construction module: Based on the feature dimensions contained in the human feature set, the profile dimension level is set, and according to the correlation mapping relationship between the profile dimension level and each feature dimension in the human feature set, a multi-layer needs profile of elderly people with dementia is constructed. Dynamic Tracing and Iteration Module: Acquires real-time monitoring data of elderly people with dementia, performs dynamic state recognition based on real-time monitoring data and multi-layered needs profiles of elderly people with dementia, and outputs profile state features; Based on the characteristics of the profile, the causes of symptoms corresponding to the multi-layered needs profile of elderly people with dementia are analyzed to obtain the causal analysis results. Decision generation and distribution module: Based on the causal tracing results, it selects an appropriate care plan from the care strategy library, performs personalized adjustments on the appropriate care plan to generate adaptive care decisions, and distributes the adaptive care decisions to various receiving terminals for storage and collaborative interaction.

2. The adaptive care decision-making system for dementia-affected elderly based on dynamic needs profiling as described in claim 1, characterized in that, The methods for obtaining the original human dataset include: The wearable sensor array deployed on the body surface of elderly people with dementia continuously collects multi-source vital signs signals, including heart rate variability sequences, body surface temperature sequences, skin conductance sequences, and limb acceleration sequences. At the same time, spatial location signals are obtained through a personal positioning tag, which includes the real-time coordinates and movement speed of the elderly people with dementia in the care area. Multi-source vital signs signals are separated into channels according to the acquisition channel identifier. Signal continuity detection is performed on the vital signs signals of each channel to identify signal interruption segments caused by sensor detachment and signal distortion segments caused by electromagnetic interference. The identified signal interruption segments and signal distortion segments are marked as invalid signal intervals and removed. Position drift detection is performed on the spatial position signal. By comparing the coordinate change at adjacent sampling times with the physiological movement limit speed of the elderly with dementia, coordinate jumps that exceed the physiological movement limit speed are marked as positioning anomalies and removed. The processed multi-source vital signs and spatial location signals are time-series arranged according to a unified timestamp to output the original human dataset.

3. The adaptive care decision-making system for dementia-affected elderly based on dynamic needs profiling as described in claim 2, characterized in that, Methods for obtaining a set of character traits include: Heart rate variation sequences were extracted from the original human dataset. The heart rate variation sequences were segmented and statistically analyzed according to the diurnal biological clock cycle. The temporal distribution of the heart rate peak-to-trough difference, the heart rate change rate, and the heart rate baseline offset between adjacent diurnal cycles were calculated within each diurnal cycle. The temporal distribution of the heart rate peak-to-trough difference, the heart rate change rate, and the heart rate baseline offset were combined to form physiological rhythm features. The movement trajectory sequence corresponding to the spatial location signal is extracted from the original human dataset. The movement trajectory sequence is subjected to residence area clustering operation to identify the high-frequency residence area and low-frequency visit area of ​​the elderly with dementia in the care area. The degree of change of daytime visit order and the repetition rate of inter-region transfer path are calculated for the high-frequency residence area. The degree of change of daytime visit order and the repetition rate of inter-region transfer path are combined into cognitive decline characteristics. Limb acceleration sequences are extracted from the original human dataset. Gait cycles are segmented into limb acceleration sequences. Step frequency stability, stride symmetry, and gait start-stop response delay are extracted for each gait cycle. Step frequency stability, stride symmetry, and gait start-stop response delay are combined into behavioral trajectory features. Skin conductance sequences were extracted from the original human dataset. The conductance response amplitude and conductance recovery time constant of the skin conductance sequences under different activity scenarios were calculated. The conductance response amplitude and conductance recovery time constant were combined into emotional fluctuation features. Physiological rhythm characteristics, cognitive decline characteristics, behavioral trajectory characteristics, and emotional fluctuation characteristics are structured and encapsulated according to feature dimension labels, and integrated into a set of personality characteristics.

4. The adaptive care decision-making system for dementia-affected elderly based on dynamic needs profiling as described in claim 3, characterized in that, The methods for constructing a multi-layered needs profile for elderly people with dementia include: For the physiological rhythm features, cognitive decline features, behavioral trajectory features and emotional fluctuation features contained in the human feature set, corresponding profile dimension levels are set respectively. The profile dimension levels include physiological baseline layer, cognitive state layer, behavioral pattern layer and emotional representation layer. The physiological baseline layer corresponds to physiological rhythm features, the cognitive state layer corresponds to cognitive decline features, the behavioral pattern layer corresponds to behavioral trajectory features, and the emotional representation layer corresponds to emotional fluctuation features. In the physiological baseline layer, the heart rate baseline offset in the physiological rhythm characteristics is used as the anchoring parameter to establish an individualized physiological baseline interval for elderly people with dementia. The upper and lower bounds of the individualized physiological baseline interval are determined by the steady-state distribution range of the heart rate peak-to-valley difference in multiple consecutive diurnal cycles of elderly people with dementia. In the cognitive state layer, the degree of change in daytime visit order and the repetition rate of inter-regional transfer paths in the cognitive decline characteristics are used as evaluation indicators to label the current cognitive function stage of elderly people with dementia. In the behavioral pattern layer, step frequency stability and stride symmetry in behavioral trajectory features are used as parameters to establish an individualized behavioral baseline map of elderly people with dementia. In the emotional representation layer, the distribution map of the emotional ground state of elderly people with dementia is established using the conductance response amplitude and conductance recovery time constant in the emotional fluctuation characteristics as parameters. By assembling individualized physiological baseline intervals, cognitive function stage labels, individualized behavioral benchmark maps, and emotional base state distribution maps in a hierarchical nested structure, a multi-layered needs profile of elderly people with dementia is constructed.

5. The adaptive care decision-making system for dementia-affected elderly based on dynamic needs profiling according to claim 4, characterized in that, Dynamic state recognition is performed based on real-time monitoring data and multi-layered needs profiles of elderly people with dementia, outputting profile state features, including: Real-time monitoring data is obtained through a wearable sensor array carried by elderly people with dementia. The real-time monitoring data includes real-time heart rate, real-time body surface temperature, real-time skin conductivity, real-time limb acceleration, and real-time spatial coordinates within the current collection period. The instantaneous heart rate value is compared with the individualized physiological baseline interval recorded in the physiological baseline layer of the multi-layered needs profile of elderly people with dementia to determine the interval assignment. When the instantaneous heart rate value falls outside the individualized physiological baseline interval, the deviation between the instantaneous heart rate value and the nearest boundary of the individualized physiological baseline interval is calculated to generate physiological deviation indicators. The real-time spatial coordinates are matched with the individualized behavioral baseline map recorded in the behavioral pattern layer of the multi-layered needs profile of elderly people with dementia. The path deviation between the current movement path and the historical regular path in the individualized behavioral baseline map is extracted to generate behavioral deviation indicators. The instantaneous skin conductance value is compared with the emotional ground state distribution map recorded in the emotional representation layer of the multi-layered needs profile of elderly people with dementia. The degree of deviation between the current conductance response and the ground state conductance range in the emotional ground state distribution map is calculated to generate an emotional deviation indicator. Physiological deviation indicators, behavioral deviation indicators, and emotional deviation indicators are structured and arranged according to indicator categories to output profile status features.

6. The adaptive care decision-making system for dementia-affected elderly based on dynamic needs profiling as described in claim 5, characterized in that, Based on the characteristics of the dementia profile, a causal analysis was conducted on the symptoms and triggering factors corresponding to the multi-layered needs profile of elderly people with dementia. The causal analysis results include: Physiological deviation indicators, behavioral deviation indicators, and emotional deviation indicators are extracted from the portrait status features. The temporal co-occurrence relationship among physiological deviation indicators, behavioral deviation indicators, and emotional deviation indicators is analyzed to identify the indicator combinations that deviate simultaneously within the same time period and mark the indicator combinations as linked deviation patterns. The linkage deviation pattern is matched with the preset knowledge graph of dementia symptom causes to determine the candidate symptom causes corresponding to the linkage deviation pattern; The candidate symptom triggers are combined with the cognitive function stage labels recorded in the cognitive state layer of the multi-layered needs profile of elderly people with dementia to perform stage adaptability verification. It is determined whether the candidate symptom triggers are consistent with the typical pathological manifestations of the current cognitive function stage of the elderly people with dementia. The candidate symptom triggers that pass the adaptability verification are identified as the causal source results.

7. The adaptive care decision-making system for dementia-affected elderly based on dynamic needs profiling as described in claim 6, characterized in that, Based on the results of causal analysis, appropriate care plans are selected from the care strategy library, including: The identified symptom trigger category identifiers and corresponding severity levels of indicators are extracted from the causal tracing results. The severity levels of indicators are graded and labeled based on the proportional relationship between the deviation of each indicator in the profile status features and the corresponding baseline range in the multi-layered needs profile of the elderly with dementia. The primary index is based on the category of symptom triggers, and the secondary index is based on the severity level of the indications. A dual search operation is performed in the hierarchical index structure of the care strategy library. The care strategy library is divided into primary partitions according to the category of symptom triggers, and each primary partition is divided into secondary partitions according to the severity level of the indications. Each secondary partition stores care plan entries under the corresponding conditions. From the set of care plan entries returned by the dual search operation, the applicable cognitive stage range field of each care plan entry is extracted. The applicable cognitive stage range field is matched with the cognitive function stage label recorded in the cognitive state layer of the multi-layered needs profile of elderly people with dementia. Care plan entries that are incompatible with the applicable cognitive stage range and cognitive function stage label are filtered out. The care plan entries that are retained after filtering are determined as suitable care plans.

8. The adaptive care decision-making system for dementia-affected elderly based on dynamic needs profiling according to claim 7, characterized in that, Personalized adjustments are made to the adapted care plan to generate adaptive care decisions, including: The care intervention action sequence and care execution time parameter are extracted from the adapted care plan. The care intervention action sequence contains multiple care intervention action nodes arranged in the order of execution. The care execution time parameter contains the suggested execution time and suggested duration of each care intervention action node. The daily activity time sequence pattern of elderly people with dementia is extracted from the individualized behavioral benchmark map of the behavior pattern layer in the multi-layered needs profile of elderly people with dementia. The daily activity time sequence pattern records the typical activity types and activity duration intervals of elderly people with dementia at different times. The suggested execution time of each care intervention action node in the care execution time parameter is checked for time period conflict with the daily activity time sequence pattern. When the suggested execution time of a care intervention action node falls into the high activity density period of the elderly person with dementia, the execution time of the care intervention action node is moved to the low activity density period adjacent to the high activity density period. From the emotional base state distribution map of the emotional representation layer in the multi-layered needs profile of elderly people with dementia, the emotional sensitive period markers of elderly people with dementia are extracted. The intervention intensity of care intervention action nodes that fall within the emotionally sensitive period is adjusted to be moderated, and the execution range of care intervention action nodes involving environmental changes is reduced. The care intervention sequence, after time period conflict adjustment and intervention intensity easing adjustment, is re-encapsulated with the corresponding care execution time parameters to generate adaptive care decisions.

9. The adaptive care decision-making system for dementia-affected elderly based on dynamic needs profiling as described in claim 8, characterized in that, Adaptive care decisions are distributed to multiple receiving terminals for storage and collaborative interaction, including: Terminal adaptability analysis is performed on adaptive care decisions, and the execution subject category identifiers of each care intervention action node in the adaptive care decision are extracted. The execution subject category identifiers are divided into caregiver execution, smart device execution, and remote medical care execution. Based on the execution subject category identifiers, the adaptive care decision is split into caregiver instruction packages for caregiver terminals, device control instruction packages for smart device terminals, and medical care notification instruction packages for remote medical care terminals. Terminal routing identifiers and instruction priority tags are attached to caregiver instruction packages, equipment control instruction packages, and medical notification instruction packages, respectively. The terminal routing identifier points to the communication address of the corresponding receiving terminal, and the instruction priority tag is determined based on the severity level of the indicators of each care intervention action node in adaptive care decision-making. The care information distribution channel pushes caregiver instruction packages, equipment control instruction packages, and medical notification instruction packages to the corresponding receiving terminals in the order of priority of instruction priority tags. After receiving the corresponding instruction package, each receiving terminal performs local storage operation and sends a reception confirmation signal back to the care information distribution channel. A collaborative interaction status table is established based on the reception confirmation signals returned by each receiving terminal. The collaborative interaction status table records the reception status, execution progress, and completion flag of the corresponding instruction packet of each receiving terminal. When the execution progress of any receiving terminal changes, a progress change notification is synchronously pushed to the other relevant receiving terminals through the collaborative interaction status table.

10. The adaptive care decision-making system for dementia-affected elderly based on dynamic needs profiling according to claim 9, characterized in that, The association deviation pattern is matched with a pre-defined knowledge graph of dementia symptom triggers to identify candidate symptom triggers corresponding to the association deviation pattern, including: Extract the combination of indicator categories that deviate simultaneously from the linkage deviation pattern and the deviation direction label of each indicator. The deviation direction label is divided into positive deviation and negative deviation. Combine the indicator category combination and the deviation direction label of each indicator into a linkage feature vector. Each symptom trigger entry is read one by one from the knowledge graph of dementia symptom triggers. Each symptom trigger entry contains the standard indicator linkage feature corresponding to the symptom trigger. The standard indicator linkage feature records the type of indicator that should deviate when the symptom trigger is triggered and the corresponding deviation direction. The linkage feature vector is compared with the standard indicator linkage feature of each symptom cause entry. The comparison operation includes determining whether the indicator category in the linkage feature vector is covered by the indicator category of the standard indicator linkage feature, and whether the deviation direction marker of each indicator in the linkage feature vector is consistent with the deviation direction of the corresponding indicator in the standard indicator linkage feature. Symptom cause entries that meet the preset matching threshold for both indicator category coverage and deviation direction consistency are identified as candidate symptom causes.

Images