Old age chronic disease body digital monitoring and intervention system based on multi-modal perception

Abstract

The application discloses a multi-modal perception-based digital monitoring and intervention system for old people with chronic diseases, and relates to the technical field of health monitoring.The application comprises the following steps: removing artifacts from physiological signals and motion signals, and using a long short-term memory network to construct a personalized physiological baseline model to generate a physiological abnormality deviation index; extracting facial key points from the face image and calculating a psychological state abnormality index, extracting an emotional distribution from the voice signal and calculating a psychological state abnormality index; calculating a mind-body coupling risk value and a cumulative risk index based on the physiological abnormality deviation index and the psychological state abnormality index; and when the cumulative risk index exceeds a preset threshold, outputting a hierarchical intervention strategy in a retrieval-enhanced generation mode based on a pre-constructed double knowledge graph.The application realizes the functional extension from passive monitoring to active intervention through the fusion analysis of multi-modal physiological and psychological data and the quantification of mind-body coupling risk.

Description

Technical Field

[0001] This invention belongs to the field of health monitoring technology, and in particular relates to an embodied digital monitoring and intervention system for chronic diseases in the elderly based on multimodal perception. Background Technology

[0002] In the field of chronic disease management in the elderly, clinical medical research has confirmed a pathological mechanism known as psychosomatic comorbidity, which establishes a bidirectional and synergistic link between chronic physical diseases such as cardiovascular disease and diabetes and psychological disorders such as anxiety and depression. Specifically, the physical discomfort and functional limitations caused by chronic diseases can affect the mood regulation center through neuroendocrine pathways, leading to increased levels of psychological stress; while persistent psychological stress can cause autonomic nervous system dysfunction and the release of inflammatory factors, in turn exacerbating the progression of the original physical diseases. This closed-loop characteristic of the interaction between mind and body makes it difficult for monitoring methods that only focus on physiological indicators or only assess psychological state to fully reflect the true health risks of the elderly with chronic diseases.

[0003] To acquire the aforementioned information on both physical and mental states, existing technologies already employ various mature data acquisition methods. At the physiological monitoring level, wearable devices equipped with photoplethysmography (PPG) sensors can continuously record vital signs such as heart rate and blood oxygen saturation at high sampling frequencies. Simultaneously, built-in accelerometers can collect three-axis motion signals to identify and eliminate motion artifacts caused by daily activities on physiological waveforms. At the psychological state perception level, audio-visual interaction devices based on consumer-grade cameras and microphone arrays possess basic capabilities for facial posture analysis and voice emotion recognition: by tracking the geometric positions of key facial points and their inter-frame displacements, morphological features related to emotional expression can be extracted; by performing acoustic parameterization processing such as Mel-frequency cepstral coefficients on the speech stream and inputting it into a lightweight neural network classifier, audio segments can be mapped into probability distribution vectors of different emotional categories. These technological components provide pathways for raw data acquisition and preliminary feature generation from both physiological and psychological dimensions.

[0004] However, when facing the monitoring needs of elderly people with chronic diseases for psychosomatic comorbidity, existing technological systems operate with physiological signal processing and psychological feature extraction links running independently, lacking a time-aligned and coupled quantitative analysis channel between the two. In actual deployments, heart rate sequences collected by wearable devices are typically only used to trigger physiological abnormality alarms based on fixed thresholds, while facial fatigue indicators and voice emotion distribution output by the audio-visual interaction module are limited to auxiliary functions for optimizing human-computer interaction experience; the two do not intersect at the data flow level. This fragmented processing architecture prevents the system from perceiving and measuring the dynamic interaction between physiological deviations amplifying psychological stress and psychological stress exacerbating physiological deviations in the psychosomatic comorbidity mechanism. Consequently, even when physiological indicators show slight deviations but the psychological state has already shown a continuous deterioration trend, the overall risk assessment remains at a low level determined by a single-dimensional criterion, missing the window of opportunity for early intervention in the early stages of psychosomatic comorbidity deterioration. Therefore, the following solutions are proposed to address these issues. Summary of the Invention

[0005] The purpose of this invention is to provide an embodied digital monitoring and intervention system for chronic diseases in the elderly based on multimodal perception. Through the fusion analysis of multimodal physiological and psychological data and the quantification of mind-body coupling risks, it realizes the functional extension from passive monitoring to active intervention, and solves the problem of the separation of mind-body data in existing monitoring programs.

[0006] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0007] This invention relates to an embodied digital monitoring and intervention system for chronic diseases in the elderly based on multimodal perception, comprising:

[0008] Smart wearable devices are used to continuously collect users' physiological and motion signals;

[0009] Smart terminals are used to collect users' facial images and voice signals;

[0010] The smart terminal is configured with an edge computing unit and performs the following steps:

[0011] Artifact removal is performed on the physiological and motion signals, and a personalized physiological baseline model is constructed using a long short-term memory network to generate a physiological abnormality deviation index.

[0012] Facial key points are extracted from the facial image and psychological state abnormality index is calculated; emotional distribution is extracted from the speech signal and psychological state abnormality index is calculated.

[0013] The psychosomatic coupling risk value and cumulative risk index are calculated based on the aforementioned physiological abnormality deviation index and psychological state abnormality index.

[0014] When the cumulative risk index exceeds a preset threshold, a graded intervention strategy is output based on a pre-constructed dual knowledge graph through a retrieval-enhanced generation method.

[0015] Furthermore, the smart wearable device includes a photoplethysmography (PPG) sensor and a triaxial accelerometer, the physiological signals include heart rate and blood oxygen saturation, and the motion signals are used to calculate motion intensity indicators to mark motion interference segments.

[0016] Furthermore, the personalized physiological baseline model is implemented through a two-layer long short-term memory network. The network outputs a baseline prediction value and an uncertainty estimate based on the current physiological feature vector and the preceding hidden state. The physiological abnormality deviation index is the Mahalanobis distance between the current physiological feature vector and the baseline prediction value.

[0017] Furthermore, the calculation of the abnormal psychological state index includes: comparing the extracted facial key points with the resting state reference coordinates to obtain a facial fatigue and anxiety composite index; inputting the Mel-frequency cepstral coefficients of the speech signal into an emotion classifier to obtain an emotion probability distribution; and calculating the KL divergence between the emotion probability distribution and the resting state reference distribution as the abnormal psychological state index.

[0018] Furthermore, the aforementioned psychosomatic coupling risk value Calculate using the following formula:

[0019]

[0020] In the formula, This is a physiological abnormality deviation index. This is an index of abnormal psychological state. This is a psychological amplification factor determined based on the user's chronic disease type. Physiological saturation inhibitor

[0021] Furthermore, the cumulative risk index Calculate using the following formula:

[0022]

[0023] In the formula, As of the date The cumulative risk index for each time window; This is the forgetting factor, used to control the rate at which historical risk values ​​decay with the current assessment.

[0024] Furthermore, the dual knowledge graph includes an elderly mental health intervention graph and a traditional Chinese medicine (TCM) disease prevention and care graph. The elderly mental health intervention graph uses emotional state, intervention techniques, and intervention resources as nodes, while the TCM disease prevention and care graph uses constitution type, care principles, and intervention plans as nodes.

[0025] Furthermore, the enhanced retrieval generation method includes: performing rule-based subgraph matching and vector similarity retrieval in the dual knowledge graph based on the current physiological and psychological characteristics, concatenating the retrieved knowledge fragments with the user's basic profile, and inputting them into a lightweight large language model to generate a tiered intervention instruction sequence.

[0026] Furthermore, the edge computing unit is also used to calculate the intervention response index after the intervention is executed. When the intervention response is lower than a preset threshold, the retrieval parameters generated by the retrieval enhancement are adjusted to achieve online adaptive optimization.

[0027] The present invention has the following beneficial effects:

[0028] This invention establishes a joint assessment framework for the psychosomatic comorbidity mechanism in elderly people with chronic diseases by synchronously collecting and fusing multimodal physiological, facial posture, and voice emotion data. The adaptive physiological baseline modeling based on LSTM eliminates the interference of individual differences on anomaly detection, and the nonlinear psychosomatic coupling risk quantification model characterizes the interactive amplification effect between physiological deviations and psychological states. The dual knowledge graph-driven retrieval enhancement generation mechanism transforms professional intervention knowledge into age-appropriate graded intervention strategies, realizing the functional extension from passive monitoring to active intervention. The overall method has edge deployment capability and takes into account both the real-time data processing and user privacy protection requirements.

[0029] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0030] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of the structure of the embodied digital monitoring and intervention system for chronic diseases in the elderly based on multimodal perception, as described in this invention. Detailed Implementation

[0032] 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.

[0033] Please see Figure 1As shown, this invention is an embodied digital monitoring and intervention system for chronic diseases in the elderly based on multimodal perception, comprising:

[0034] Smart wearable devices are used to continuously collect users' physiological and motion signals;

[0035] Smart terminals are used to collect users' facial images and voice signals;

[0036] Configure the edge computing unit on the smart terminal and perform the following steps:

[0037] Artifact removal is performed on physiological and motor signals, and a personalized physiological baseline model is constructed using a long short-term memory network to generate a physiological abnormality deviation index.

[0038] Facial key points are extracted from facial images and psychological state abnormality index is calculated; emotional distribution is extracted from speech signals and psychological state abnormality index is calculated.

[0039] The psychosomatic coupling risk value and cumulative risk index are calculated based on the physiological abnormality deviation index and the psychological state abnormality index.

[0040] When the cumulative risk index exceeds a preset threshold, a tiered intervention strategy is output based on a pre-built dual knowledge graph through a retrieval-enhanced generation method.

[0041] The smart wearable device includes a photoplethysmography (PPG) sensor and a triaxial accelerometer. Physiological signals include heart rate and blood oxygen saturation, while motion signals are used to calculate exercise intensity indicators to mark motion interference segments.

[0042] The personalized physiological baseline model is implemented through a two-layer long short-term memory network. The network outputs the baseline prediction value and uncertainty estimate based on the current physiological feature vector and the previous hidden state. The physiological abnormality deviation index is the Mahalanobis distance between the current physiological feature vector and the baseline prediction value.

[0043] The calculation of the abnormal psychological state index includes: comparing extracted facial key points with resting state reference coordinates to obtain a composite index of facial fatigue and anxiety; inputting the Mel-frequency cepstral coefficients of the speech signal into an emotion classifier to obtain the emotion probability distribution; and calculating the KL divergence between the emotion probability distribution and the resting state reference distribution as the abnormal psychological state index.

[0044] Psychosomatic coupling risk value Calculate using the following formula:

[0045]

[0046] In the formula, This is a physiological abnormality deviation index. This is an index of abnormal psychological state. This is a psychological amplification factor determined based on the user's chronic disease type. Physiological saturation inhibitor

[0047] Cumulative Risk Index Calculate using the following formula:

[0048]

[0049] In the formula, As of the date The cumulative risk index for each time window; This is the forgetting factor, used to control the rate at which historical risk values ​​decay with the current assessment.

[0050] The dual knowledge graphs include an intervention graph for elderly mental health and a graph for TCM disease prevention and care. The intervention graph for elderly mental health uses emotional state, intervention techniques and intervention resources as nodes, while the TCM disease prevention and care graph uses constitution type, care principles and intervention plans as nodes.

[0051] The enhanced retrieval generation method includes: performing rule-based subgraph matching and vector similarity retrieval in the dual knowledge graphs based on current physiological and psychological characteristics, concatenating the retrieved knowledge fragments with the user's basic profile, and inputting them into a lightweight large language model to generate a tiered intervention instruction sequence.

[0052] The edge computing unit is also used to calculate the intervention response rate index after the intervention is implemented. When the intervention response rate is lower than the preset threshold, the retrieval parameters generated by the retrieval enhancement are adjusted to achieve online adaptive optimization.

[0053] The specific application of this embodiment is as follows:

[0054] Step S1: Synchronous acquisition and spatiotemporal alignment of multimodal data

[0055] A monitoring system consisting of smart terminals and smart wearable devices is deployed in a home setting. The smart terminal is a device with a display screen, camera, microphone array and edge computing capabilities, such as a tablet computing device. The smart terminal is used to perform audio and video data acquisition, local data processing and human-computer interaction. The smart wearable device is a smart bracelet or smartwatch with a photoplethysmography (PPG) sensor and a triaxial accelerometer, worn on the user's wrist or finger, for continuous collection of physiological signals.

[0056] Smart wearable devices at the first sampling frequency Continuously collect users' heart rate data and blood oxygen saturation data ,in Sampling time, The value range is 25Hz to 100Hz; meanwhile, the accelerometer built into the wearable device operates at a second sampling frequency. Acquisition of three-axis motion signals , , , used for motion artifact labeling;

[0057] The smart terminal uses a built-in high-definition camera at a third sampling frequency Periodically capture user facial image sequences ,in Indicates the frame number. The sampling rate is set to 1 to 5 frames per minute; the smart terminal's microphone array continuously acquires ambient audio streams at a sampling rate of 16kHz, and after voice activity detection, extracts valid speech segments from the user. ;

[0058] To achieve spatiotemporal alignment of multimodal data, the system uses the local clock of the smart terminal as the reference timestamp and performs linear interpolation resampling on the physiological data stream from the wearable device and the audio and video data stream from the smart terminal itself to generate a unified time granularity. Multimodal synchronization data matrix ,in For the first Physiological feature vectors within a time window This is a facial feature vector. For speech emotion feature vectors, Set to 1 minute; physiological feature vector It should at least include a set of mean heart rate, time-domain indicators of heart rate variability, and mean blood oxygen saturation within that time window; facial posture feature vectors. Includes body posture features derived from facial key point coordinates; and voice emotion feature vectors. It contains emotion distribution information derived from acoustic features.

[0059] Step S2: Adaptive physiological baseline modeling and abnormal deviation quantification based on LSTM

[0060] In view of the significant individual differences in physiological indicators among the elderly with chronic diseases and the interference of daily activities, this invention uses a long short-term memory network to establish a personalized physiological baseline model for each user.

[0061] Step S21: Process the physiological feature vector sequence obtained in step S1 Perform motion artifact removal; calculate the first... Exercise intensity index within a time window :

[0062]

[0063] In the formula, For the first Exercise intensity index within a time window; For the first The total number of triaxial acceleration sampling points collected by the accelerometer of the smart wearable device within a time window; This is the index variable for the sampling time. For the first All sampling times within a time window The set that constitutes; To be at the sampling time At that time, the accelerometer was in The acceleration component measured along the axial direction; To be at the sampling time At that time, the accelerometer was in The acceleration component measured along the axial direction; To be at the sampling time At that time, the accelerometer was in The acceleration component measured along the axial direction; when At that time, the physiological data in this window is marked as a motion interference segment and is not included in the baseline update. The preset exercise intensity threshold;

[0064] Step S22: Input the artifact-filtered resting-state physiological feature sequence into the LSTM baseline tracking network; this network consists of two stacked LSTM units, with a hidden layer dimension of 64; in the... Each time window, the network's hidden state is based on the preceding window. and current input Output baseline prediction value and uncertainty estimation :

[0065]

[0066] In the formula, For LSTM networks targeting the first The baseline value vector of physiological characteristics predicted by each time window; For LSTM networks targeting the first The estimated uncertainty of the forecast output for each time window; This is the forward propagation function of a recurrent neural network consisting of two stacked layers of long short-term memory units; For the first Within a time window, the resting-state physiological feature input vector after motion artifact removal; For LSTM networks in processing the first The hidden state vector output after a time window; It is the set of all trainable weights in the LSTM network;

[0067] Step S23: Define the first Physiological abnormality deviation index within a time window :

[0068]

[0069] In the formula, It is a diagonal covariance matrix; In essence, it is the Mahalanobis distance, and the larger the value, the more significant the deviation of the current physiological state from the personalized baseline.

[0070] Step S3: Feature extraction of facial posture and vocal emotion, and quantification of psychological state.

[0071] Step S31: Process the facial image sequence obtained in step S1 This invention employs a lightweight facial landmark detection network to extract the two-dimensional coordinates of 468 facial feature points. Addressing the issue of facial muscle relaxation leading to decreased facial expression amplitude in the elderly, this invention defines a composite index of facial fatigue and anxiety based on the keypoint displacement ratio. :

[0072]

[0073] In the formula, For the first The first window Coordinate vectors of key points around the eyes and corners of the mouth; The mean of the resting state reference coordinates collected during the system initialization phase for this user; The number of key points selected; Index variables for key points; The rate of change in eyelid opening and closing; , These are the combination coefficients obtained through pre-trained regression;

[0074] Step S32: Process the speech segment Mel frequency cepstral coefficients are extracted as acoustic features, input to a lightweight Transformer-based sentiment classifier, and output a four-dimensional sentiment probability distribution vector. Take the first The mean of the emotion vectors of all speech segments within a given time window is used as the speech emotion feature vector for that window. ;

[0075] Step S33: Define the first Abnormal psychological state index within a time window The KL divergence between the sentiment distribution and the reference baseline distribution:

[0076]

[0077] In the formula, To reference sentiment distribution Current sentiment distribution Kullback–Leibler divergence function; This is an index variable for the sentiment category; The reference emotion distribution vector for the calm state collected by the user during the system initialization phase; and These are the current window and the reference distribution, respectively. The probability value of a type of emotion.

[0078] Step S4: Joint Quantification and Time-Series Cumulative Assessment of Psychosomatic Coupling Risk

[0079] Step S41: Define the first Risk value of psychosomatic coupling within a time window :

[0080]

[0081] In the formula, The psychological amplification factor is determined by looking up a table based on the user's chronic disease type and duration of medical history. This is a physiological saturation inhibitory factor used to simulate the diminishing marginal effect of psychological impact when there are severe physiological abnormalities; the physical meaning of this formula is: when the degree of deviation from physiological abnormality... When the condition is relatively mild, the deterioration of the psychological state will exponentially amplify the risk of psychosomatic comorbidity; when the physiological abnormalities are already at a high level, the additional amplifying effect of psychological factors tends to saturate, which is consistent with the clinical observation pattern of psychosomatic comorbidity.

[0082] Step S42: Considering the time-series cumulative effect of psychosomatic risks, define the cumulative risk index. :

[0083]

[0084] In the formula, As a forgetting factor, it controls the decay rate of historical risk values ​​on the current assessment; the introduction of an accumulation mechanism can effectively filter out single-point noise interference and accurately capture the continuous deterioration trend of mental and physical state.

[0085] Step S5: Generation of RAG intervention strategies driven by dual knowledge graphs

[0086] When the cumulative risk index When the preset grading threshold is exceeded, the system triggers the corresponding level of proactive intervention process; the intervention strategy is generated using a dual knowledge graph collaborative retrieval enhancement generation framework.

[0087] The first knowledge graph is a graph for interventions in elderly mental health. Its nodes include emotional states, cognitive behavioral therapy techniques, music intervention repertoire, breathing training programs, etc., and the edges represent applicable relationships and contraindications; the second knowledge graph is a TCM disease prevention and care graph. Its nodes include constitution identification types, the principle of emotional balance, the correspondence between the five elements and music, and suggestions for acupoint massage, while the edges represent the relationships of mutual generation and restraint and intervention compatibility;

[0088] Physiological features extracted for the current time window With psychological characteristics First, in the map The system performs rule-based subgraph matching to retrieve a care plan template corresponding to the current symptom characteristics; subsequently, it performs subgraph matching within the graph. China-Israel sentiment distribution vector Using vector similarity retrieval as the query criteria, the top results are retrieved. One intervention knowledge item;

[0089] The retrieved structured knowledge fragments are concatenated with the user's basic profile information to form contextual prompts, which are then input into a lightweight large language model deployed at the edge of a smart terminal. This model outputs natural language intervention instructions based on a retrieval-enhanced generation paradigm, including tiered intervention action sequences. For mild risks, it outputs music push instructions and guided breathing training scripts. For moderate risks, it outputs a structured dialogue tree based on a simplified version of cognitive behavioral therapy. For severe risks, it generates a risk briefing and pushes it to the family terminal via a message queue.

[0090] The entire process of generating the above intervention is completed locally on the smart terminal. Only the anonymized statistical features after desensitization are uploaded to the cloud for offline model iteration, which meets the requirements of privacy protection and compliance.

[0091] Step S6: Intervention effect feedback and online model adaptation

[0092] The system on the day after the intervention is executed Physiological and psychological data are continuously collected within a tracking window to calculate intervention response indicators. :

[0093]

[0094] In the formula, The cumulative risk index at the moment of intervention triggering; After the intervention ended, the first The cumulative risk index for each window;

[0095] like If the value falls below the preset improvement threshold, the system will adjust the similarity threshold or intervention intensity level parameter of the RAG retrieval in step S5 to achieve online adaptive optimization of the intervention strategy.

[0096] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0097] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A multimodal perception-based embodied digital monitoring and intervention system for chronic diseases in the elderly, characterized in that, The method includes: Smart wearable devices are used to continuously collect users' physiological and motion signals; Smart terminals are used to collect users' facial images and voice signals; The smart terminal is configured with an edge computing unit and performs the following steps: Artifact removal is performed on the physiological and motion signals, and a personalized physiological baseline model is constructed using a long short-term memory network to generate a physiological abnormality deviation index. Facial key points are extracted from the facial image and psychological state abnormality index is calculated; emotional distribution is extracted from the speech signal and psychological state abnormality index is calculated. The psychosomatic coupling risk value and cumulative risk index are calculated based on the aforementioned physiological abnormality deviation index and psychological state abnormality index. When the cumulative risk index exceeds a preset threshold, a graded intervention strategy is output based on a pre-constructed dual knowledge graph through a retrieval-enhanced generation method.

2. The embodied digital monitoring and intervention system for chronic diseases in the elderly based on multimodal perception as described in claim 1, characterized in that, The smart wearable device includes a photoplethysmography (PPG) sensor and a triaxial accelerometer. The physiological signals include heart rate and blood oxygen saturation. The motion signals are used to calculate motion intensity indicators to mark motion interference segments.

3. The embodied digital monitoring and intervention system for chronic diseases in the elderly based on multimodal perception as described in claim 1, characterized in that, The personalized physiological baseline model is implemented through a two-layer long short-term memory network. The network outputs the baseline prediction value and uncertainty estimate based on the current physiological feature vector and the previous hidden state. The physiological abnormality deviation index is the Mahalanobis distance between the current physiological feature vector and the baseline prediction value.

4. The embodied digital monitoring and intervention system for chronic diseases in the elderly based on multimodal perception as described in claim 1, characterized in that, The calculation of the abnormal psychological state index includes: comparing the extracted facial key points with the resting state reference coordinates to obtain a facial fatigue and anxiety composite index; inputting the Mel frequency cepstral coefficients of the speech signal into an emotion classifier to obtain an emotion probability distribution; and calculating the KL divergence between the emotion probability distribution and the resting state reference distribution as the abnormal psychological state index.

5. The embodied digital monitoring and intervention system for chronic diseases in the elderly based on multimodal perception according to claim 1, characterized in that, The risk value of psychosomatic coupling Calculate using the following formula: In the formula, This is a physiological abnormality deviation index. This is an index of abnormal psychological state. This is a psychological amplification factor determined based on the user's chronic disease type. It is a physiological saturation inhibitor.

6. The embodied digital monitoring and intervention system for chronic diseases in the elderly based on multimodal perception according to claim 5, characterized in that, The cumulative risk index Calculate using the following formula: In the formula, As of the date The cumulative risk index for each time window; This is the forgetting factor, used to control the rate at which historical risk values ​​decay with the current assessment.

7. The embodied digital monitoring and intervention system for chronic diseases in the elderly based on multimodal perception according to claim 1, characterized in that, The dual knowledge graph includes an elderly mental health intervention graph and a traditional Chinese medicine (TCM) disease prevention and care graph. The elderly mental health intervention graph uses emotional state, intervention techniques, and intervention resources as nodes, while the TCM disease prevention and care graph uses constitution type, care principles, and intervention plans as nodes.

8. The embodied digital monitoring and intervention system for chronic diseases in the elderly based on multimodal perception according to claim 7, characterized in that, The enhanced retrieval generation method includes: performing rule-based subgraph matching and vector similarity retrieval in the dual knowledge graph based on the current physiological and psychological characteristics, concatenating the retrieved knowledge fragments with the user's basic profile, and inputting them into a lightweight large language model to generate a tiered intervention instruction sequence.

9. The embodied digital monitoring and intervention system for chronic diseases in the elderly based on multimodal perception according to claim 1, characterized in that, The edge computing unit is also used to calculate the intervention response index after the intervention is executed. When the intervention response is lower than a preset threshold, the retrieval parameters generated by the retrieval enhancement are adjusted to achieve online adaptive optimization.

Images