A Smart Elderly Care Service Management System and Method with Anomaly Alert Function

Abstract

This invention discloses a smart elderly care service management system and method with anomaly alert function, relating to the field of elderly care service technology. It includes: a data acquisition module for collecting data; a data preprocessing module for preprocessing the collected data and constructing a database specifically for the elderly; a model building and optimization module for building, training, and optimizing anomaly prediction models; an anomaly identification and early warning module for predicting and comparing real-time collected data, determining the degree of data deviation, and triggering early warnings; a smart service interaction module for pushing early warning information to service personnel, the elderly, and their families; a mindset management module for monitoring the well-being of the elderly; and a resident mutual assistance module for providing mutual assistance reminders. This invention can predict health data for future moments, classifying anomalies based on relative deviation, solving the problem of timely detection of sudden health risks in traditional elderly care, and improving the timeliness of elderly care services.

Description

Technical Field

[0001] This invention relates to the field of elderly care service technology, specifically to a smart elderly care service management system and method with anomaly alert function. Background Technology

[0002] With the accelerating aging of my country's population, the proportion of elderly people is increasing year by year, making health and elderly care a focus of widespread social concern. Currently, the elderly population faces the dilemma of unstable health conditions and difficulty in timely detection of sudden health risks. Traditional health monitoring methods are mostly passive health management models, relying on the elderly to actively measure and report their health status or regularly visit medical institutions for check-ups. This method is insufficient to meet the growing health needs and expectations of the elderly for efficient services.

[0003] Traditional health monitoring methods lack real-time and personalized monitoring tools, resulting in many elderly people not receiving timely warnings and interventions when health risks suddenly occur. Furthermore, the methods for collecting health data are limited and cannot fully reflect the vital signs and daily activities of the elderly, thus restricting the accuracy of health monitoring.

[0004] Based on this, a smart elderly care service management system and method with anomaly alert function is now provided. By utilizing technologies such as the Internet of Things and wearable devices, it is possible to collect and monitor the vital signs and home behavior of the elderly in real time, which can eliminate the drawbacks of existing technical solutions. Summary of the Invention

[0005] The purpose of this invention is to provide a smart elderly care service management system and method with anomaly alert function, in order to solve the problems of existing health monitoring methods in the background art lacking real-time and personalized monitoring means and having limited data collection methods.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A smart elderly care service management system with anomaly alert function includes: The data acquisition module is used to collect physiological, behavioral, and psychological data of the elderly. The data preprocessing module is used to store and preprocess the collected data to build a database specifically for the elderly. The model building and optimization module trains and iteratively optimizes the model based on historical data in the database specifically for the elderly, resulting in the final anomaly prediction model. The anomaly identification and early warning module uses an anomaly prediction model to predict and compare real-time collected data, determine the degree of data deviation and trigger corresponding anomaly level warnings. The anomaly prediction model is a multi-dimensional time series prediction model based on LSTM, which is used to capture the time series change patterns and inter-dimensional correlation features of physiological and behavioral data, and realize the prediction of future moments and anomaly determination of elderly health data. The intelligent service interaction module is used to push early warning information, abnormal data, abnormal analysis reports and targeted intervention measures to service personnel, the elderly and their families; The mindset management module is used to analyze the mindset data of the elderly and monitor their happiness level. The resident mutual assistance module is used to enable authorized sharing of elderly people's physical data and mutual assistance reminders among residents; The modules communicate with each other through the network to achieve collaborative work.

[0007] Furthermore, the data acquisition module includes several wearable monitoring devices and a mental health data acquisition device. The wearable monitoring devices are equipped with physiological sensors and positioning units to collect the elderly's dynamic physiological data and movement trajectory in real time. The physiological sensors include heart rate, blood pressure, blood oxygen, body temperature, and blood glucose sensors. The mental health data acquisition device uses an age-friendly interactive terminal to collect the elderly's mental health data by pushing games and psychological questions. The mini-games include memory-based and brain-training lightweight games. The psychological questions use a standardized happiness assessment scale and simultaneously collect the elderly's operation time, question and answer results, and voice emotion data during the interaction process.

[0008] Furthermore, the data preprocessing module is used to filter, deduplicate, normalize, and fill missing values ​​in the raw data collected by the data acquisition module. Specifically, it involves: using median filtering to remove noise from physiological data, using linear interpolation to fill in short-term missing data, and using the min-max normalization formula for normalization. The min-max normalization formula is expressed as: ,in, This is the original data. The minimum value of the data. The maximum value of the data is used to obtain preprocessed data after filtering, deduplication, normalization, and missing value imputation. The preprocessed data is used for model training and real-time analysis. The data preprocessing module has a built-in database specifically for the elderly, which is divided into physiological sub-database, behavioral sub-database, mental state sub-database and log sub-database according to data type.

[0009] Furthermore, the model building and optimization module includes a model building unit, a model training unit, and a model optimization unit. The model building unit combines the elderly's historical health data and individual physiological characteristics to build an initial anomaly prediction model based on LSTM, which is used to capture the temporal variation patterns of physiological and behavioral data. The model training unit uses preprocessed historical data from the elderly's exclusive database as the training set and labels the data deviations from the elderly's personalized normal baseline to supervise the training of the initial anomaly prediction model. During the training process, the gradient descent method is used to adjust the model parameters. The model optimization unit periodically collects new data and early warning response feedback data to iteratively optimize the trained anomaly prediction model. The anomaly prediction model comprises an input layer, an embedding layer, an LSTM feature extraction layer, a fully connected fusion layer, and an output layer connected in sequence. The input layer receives multi-dimensional time-series data output from the data preprocessing module, with the input being a time-series feature vector containing physiological and behavioral dimensions. The embedding layer vectorizes the discrete behavioral features of the input and concatenates them with continuous physiological features to form a feature matrix of a unified dimension. The LSTM feature extraction layer comprises multiple stacked LSTM units, capturing the long-short-term dependencies of the time-series data through a gating mechanism, and outputting a time-series feature map of the corresponding dimension. The fully connected fusion layer comprises two fully connected layers. The first fully connected layer flattens the time-series feature map into a one-dimensional feature vector and performs a non-linear transformation using the ReLU activation function. The second fully connected layer performs dimensionality reduction and fusion on the transformed features, outputting a predicted feature vector. The output layer uses a linear activation function to map the fused predicted feature vector to future... Predicted health data values ​​at any given time; The gradient descent method mentioned is the Adam adaptive gradient descent method, specifically expressed as follows: ; ; ; ; ; in, For the first First-order moment estimation This represents the current iteration number. The first-order moment attenuation coefficient, For the first First-order moment estimation For second-order moment estimation, The second-order moment attenuation coefficient, For the first Second-order moment estimation, For the loss function with respect to model parameters gradient, and This is the moment estimate after bias correction. For the updated model parameters, For the first Model parameters for the next iteration For learning rate, To prevent the minimum value where the denominator is 0; The loss function is a weighted fusion loss function of mean squared error and mean absolute error, used to measure the deviation between the model's predicted values ​​and the true values. Its expression is: ; in, For loss function, These are weighting coefficients. The mean square error and satisfying , The mean absolute error and satisfying , The number of training samples, For the first The true value of each sample For the first The model prediction value for each sample.

[0010] Furthermore, the anomaly identification and early warning module specifically includes: The real-time prediction unit uses an anomaly prediction model to predict the physiological and behavioral data collected in real time by the data acquisition module and outputs the predicted data for future moments. The deviation calculation unit is used to calculate the relative deviation between the real-time collected data, the model predicted data, and the individualized normal baseline of the elderly. The formula for calculating the relative deviation is as follows: ,in, This refers to the relative deviation. for Real-time data values ​​collected at any given moment. for The personalized normal baseline value of the elderly at any given time is obtained by the model building and optimization module based on the mean, variance and trend characteristics of the elderly’s historical health data. Anomaly grading unit is used to classify anomalies into minor deviation, moderate deviation, and severe deviation based on their relative degree of deviation; The early warning push unit pushes early warning information to the corresponding terminals according to the level of abnormality. When there is a slight deviation, it is pushed to the service personnel terminal; when there is a moderate deviation, it is pushed to the service personnel terminal, family member terminal, and resident terminal; when there is a severe deviation, it is pushed to the service personnel terminal, family member terminal, and community emergency equipment. The early warning information includes the elderly person's identity information, abnormal data, relative deviation degree, occurrence time, location information, potential risks, and handling suggestions.

[0011] Furthermore, the mindset management module specifically includes: The happiness index calculation unit, based on collected mental health data, uses a weighted summation formula to calculate the happiness index of the elderly, specifically expressed as follows: ,in, To improve the happiness index of the elderly. To score points for game interactions, Scoring based on psychological questions, To score the emotion of the voice, Scoring based on social interaction, , , and All are weighting coefficients for the corresponding terms, and satisfy the following conditions: ; The abnormal mindset recognition unit has a built-in preset happiness index threshold. When the elderly person's happiness index does not meet the preset happiness index threshold, it is judged as an abnormal mindset. The judgment is further aided by the proportion of negative emotions in the voice emotion data, and the judgment result is pushed to the service personnel terminal and the family member terminal.

[0012] Furthermore, the resident mutual assistance module specifically includes: The data authorization and sharing unit is used to set data sharing permissions and share some non-sensitive physiological data of the elderly to designated community residents' terminals. The mutual assistance reminder unit is used to push mutual assistance reminders to authorized resident terminals when the elderly person shows slight or moderate deviation, reminding residents to pay attention to the elderly person's condition; The mutual aid record unit records the initiation, response, and completion of mutual aid among residents, forming a mutual aid log which is stored in a database specifically for the elderly.

[0013] A smart elderly care service management method with anomaly alert function, applied to a smart elderly care service management system with anomaly alert function, specifically includes the following steps: Step S1: Collect the elderly’s physiological data, behavioral data and mental data in real time through several wearable monitoring devices and mental data collection devices, and transmit them to the data preprocessing module. Step S2: The collected raw data is preprocessed by the data preprocessing module, and the preprocessed data is classified and stored in the corresponding sub-databases to build a database specifically for the elderly. Step S3: Based on historical data in the database specifically for the elderly, the model building and optimization module builds an initial anomaly prediction model based on LSTM. Through supervised training, cross-validation, and iterative optimization, an anomaly prediction model that meets the prediction accuracy requirements is obtained. Step S4: Input the real-time collected data into the anomaly prediction model for prediction. The relative deviation between the real-time collected data, the model prediction data, and the elderly person's personalized normal baseline is used to classify the anomalies based on the magnitude of the relative deviation. Step S5: Based on the anomaly classification, push the early warning information to the corresponding terminal through the smart service interaction module, and generate an anomaly analysis report and targeted intervention measures. Service personnel and family members perform the handling operations based on the early warning information and targeted intervention measures and provide feedback on the results.

[0014] Furthermore, the smart service interaction module includes service personnel terminals, elderly terminals, family member terminals, and resident terminals; The service personnel terminal is used to receive early warning information, view abnormal data and analysis reports, implement targeted intervention measures, and provide feedback on the handling results; The elderly terminal uses an age-friendly interface and supports viewing real-time health data, receiving simple early warning reminders, and initiating emergency calls. The family member terminal is used to view the elderly person's real-time data, abnormal warnings, intervention reports and mutual assistance records, and to communicate with the service personnel's terminal in real time. The resident terminal is used to receive mutual assistance reminders, view authorized elderly data, and initiate mutual assistance responses, thereby achieving collaborative care among community residents.

[0015] Furthermore, the intelligent elderly care service management method also includes: Based on the collected mental state data, the happiness index of the elderly is calculated and compared with the preset happiness index threshold to identify abnormal mental states and trigger reminders. The elderly and their families can set data sharing permissions through the resident mutual assistance module, sharing non-sensitive data to designated resident terminals. When the elderly experience abnormalities, mutual assistance reminders are pushed to authorized resident terminals, and the mutual assistance process is recorded. Regularly summarize the collected data and mutual assistance records, and feed them back to the model building and optimization module to iteratively optimize the anomaly prediction model, forming a closed-loop management system.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention sets up a multi-dimensional time series prediction model based on LSTM, which can capture the time series change patterns and inter-dimensional correlation features of elderly health data, achieve the prediction effect of health data at future moments, and realize the classification of abnormality level through relative deviation, thus solving the problem of difficulty in timely detection of sudden health risks in traditional elderly care and improving the timeliness of elderly care services. 2. This invention collects mental health data from multiple dimensions to construct a happiness index for the elderly, which facilitates the identification and timely intervention of abnormal mental states, meets the mental health needs of the elderly, and achieves both physical and mental well-being in the elderly. 3. This invention integrates multiple stakeholders, including service personnel, family members, residents, and emergency medical institutions. Through the smart service interaction module, it enables data exchange and real-time communication across multiple terminals. Through the resident mutual assistance module, it enables collaborative care within the community. This solves the problem of insufficient care for elderly people living alone and those in empty nests, and improves the coverage of elderly care services. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall system structure of the present invention.

[0018] Figure 2 This is a schematic diagram of the data acquisition module of the present invention.

[0019] Figure 3 This is a schematic diagram of the data preprocessing module of the present invention.

[0020] Figure 4 This is a schematic diagram of the model construction and optimization module of the present invention.

[0021] Figure 5 This is a schematic diagram of the anomaly identification and early warning module of the present invention.

[0022] Figure 6 This is a schematic diagram of the intelligent service interaction module of the present invention.

[0023] Figure 7 This is a schematic diagram of the mindset management module of the present invention.

[0024] Figure 8 This is a schematic diagram of the resident mutual assistance module of the present invention.

[0025] Figure 9 This is a schematic diagram of the method steps of the present invention. Figure label annotations: Data acquisition module 10, wearable monitoring device 11, mentality data acquisition device 12, data preprocessing module 20, model building and optimization module 30, model building unit 31, model training unit 32, model optimization unit 33, anomaly identification and early warning module 40, real-time prediction unit 41, deviation calculation unit 42, anomaly classification unit 43, early warning push unit 44, smart service interaction module 50, service personnel terminal 51, elderly terminal 52, family member terminal 53, resident terminal 54, mentality management module 60, happiness index calculation unit 61, mentality anomaly identification unit 62, resident mutual assistance module 70, data authorization and sharing unit 71, mutual assistance reminder unit 72, mutual assistance record unit 73. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0027] Example 1 In this embodiment, as Figures 1-8 As shown, the present invention provides a smart elderly care service management system with anomaly alert function, comprising: The data acquisition module is used to collect physiological, behavioral, and psychological data of the elderly. Data preprocessing module 20 is used to store the collected data and preprocess it to build a database specifically for the elderly; The model building and optimization module 30 trains and iteratively optimizes the model based on historical data in the database specifically for the elderly, and obtains the final anomaly prediction model. The anomaly identification and early warning module 40 uses an anomaly prediction model to predict and compare real-time collected data, determine the degree of data deviation and trigger corresponding anomaly level warnings. The anomaly prediction model is a multi-dimensional time series prediction model based on LSTM, which is used to capture the time series change patterns and inter-dimensional correlation features of physiological and behavioral data, and realize the prediction of future moments and anomaly judgment of elderly health data. The intelligent service interaction module 50 is used to push early warning information, abnormal data, abnormal analysis reports and targeted intervention measures to service personnel, the elderly and their families; The mindset management module 60 is used to analyze elderly people's mindset data and monitor their happiness level. The Resident Mutual Assistance Module 70 is used to enable authorized sharing of elderly people's physical data and mutual assistance reminders among residents; The modules communicate with each other via a network, using wireless communication methods such as Wi-Fi, 4G / 5G, or Bluetooth to achieve collaborative work.

[0028] Among them, such as Figure 2As shown, the data acquisition module 10 includes several wearable monitoring devices 11 and mental health data acquisition devices 12. The wearable monitoring devices 11 are common monitoring devices on the market, such as wearable blood pressure monitors, positioning bracelets, body temperature measuring instruments, heart rate monitors, etc. The wearable monitoring devices 11 have built-in physiological sensors and positioning units for real-time collection of the elderly's dynamic physiological data and movement trajectory. The physiological sensors include, but are not limited to, heart rate, blood pressure, blood oxygen, body temperature, and blood glucose sensors. In addition to core indicators such as heart rate, blood pressure, blood oxygen, body temperature, and blood glucose, the corresponding sensors can also collect data such as step count and sleep duration. Sleep duration / sleep quality, respiratory rate, and other derived physiological data are collected. The collection frequency can be flexibly set according to the elderly person's health condition. For example: 30 minutes / time for healthy elderly people, 15 minutes / time for elderly people with chronic diseases. The mental health data collection device 12 uses an age-friendly interactive terminal. An age-friendly interactive terminal refers to a terminal device specifically designed for the physiological and psychological characteristics of the elderly. Examples include elderly-specific tablets, smart voice speakers, and other age-friendly interface devices with large fonts and icons. Mental health data of the elderly is collected by pushing games and psychological questions. The games include memory-based and brain-training lightweight games. The questions were conducted using a standardized well-being assessment scale, simultaneously collecting data on the elderly's interaction time, question-and-answer results, and emotional voice during the interaction process. Specifically, when engaging in game interaction, memory-based (e.g., number memory, picture matching) and puzzle-based (e.g., simple Sudoku, jigsaw puzzles, idiom completion) games were pushed, including but not limited to number memory, picture matching, simple Sudoku, and jigsaw puzzles. The game difficulty was divided into beginner, intermediate, and advanced levels, automatically selected according to the elderly's needs. Data collection dimensions included game completion time, accuracy rate, number of retries, and operation fluency. When conducting standardized psychological questions, the Newfoundland Memorial Questionnaire was used. Standardized well-being assessment scales specifically designed for the elderly, such as the University Happiness Scale (MUNSH) and the Geriatric Depression Scale (GDS), present questions via voice broadcast or text display. The questions are primarily multiple-choice and simple true / false questions, avoiding open-ended answers. The scale collects the elderly's responses, answering time, and number of hesitations. Simultaneously, during interactive games and psychological questioning, the scale collects the elderly's emotional voice data through the voice module built into the age-friendly interactive terminal. Voice recognition technology is used to analyze tone of voice, speech rate, and emotional keywords to determine emotional tendencies such as happiness, calmness, anxiety, and depression.

[0029] Among them, such as Figure 3 As shown, the data preprocessing module 20 is used to filter, deduplicate, normalize, and fill missing values ​​in the raw data collected by the data acquisition module 10. Specifically, it performs format standardization processing on the raw data, converting the unstructured sensor data of the wearable monitoring device 11 and the mentality data of the mentality data acquisition device 12 into a unified format, for example: timestamp + data dimension + value / label, to unify the data time unit and value unit. Median filtering is used to remove noise from physiological data. This is expressed as follows: For continuous physiological time-series data such as heart rate, blood pressure, and blood oxygen, the median of a fixed window size (e.g., 5 data points) is selected to replace the original data within the window. This operation can effectively eliminate impulse noise and random interference. For behavioral data (e.g., movement trajectory), moving average filtering is used to assist in processing to ensure the smoothness of the data. Considering that psychological data is non-continuous interactive feature data, there is no need for filtering and noise reduction. Therefore, it can directly enter the deduplication stage. The deduplication process is expressed as follows: For duplicate records with the same timestamp and the same data dimension, the value of the first collection is retained. For approximately duplicate data with a timestamp deviation within 1 second, the average value is taken as the final data. Linear interpolation is used to complete short-term missing data. This means that the filling method is adapted to different missing scenarios according to the duration and data type of missing data. If the missing data is short-term, linear interpolation is used. If the missing data is medium- to long-term, historical mean interpolation is used. If the missing data is psychological data, considering that psychological data is not frequently collected, no interpolation is performed when it is missing. It is directly marked as not collected and does not participate in the current psychological analysis. The min-max normalization formula is used to normalize all continuous numerical data (numerical features of physiological and behavioral data) after filtering, deduplication, and padding, mapping the data to the [0,1] interval to eliminate the dimensional differences between different data dimensions. The min-max normalization formula is expressed as: ,in, This is the original data. The minimum value of the data. The maximum value of the data is the preprocessed data obtained after filtering, deduplication, normalization and missing value imputation. The preprocessed data is used for model training and real-time analysis. Specifically, before operating this module, a reasonable range of values ​​for different data is set based on existing common knowledge, and an outlier screening is performed. Data that exceeds the reasonable range is marked as data to be verified and stored separately. Subsequent verification is performed to determine whether it is due to equipment acquisition error. For example, if the normal range of heart rate is set to 60-100 beats / minute, and the monitored heart rate range is 180-220 beats / minute, then an outlier screening is performed. The data preprocessing module 20 has a built-in database specifically for the elderly, which is divided into physiological sub-database, behavioral sub-database, mental state sub-database and log sub-database according to data type. Specifically, before operating this module, each elderly person is assigned a unique identification ID. All data is bound one-to-one with the elderly person's ID. The physiological sub-database stores pre-processed real-time and historical physiological data of the elderly person, including but not limited to standardized data collected by all physiological sensors such as heart rate, blood pressure, blood oxygen, body temperature, blood sugar, respiratory rate, sleep quality, and number of steps. The behavioral sub-database stores pre-processed behavioral characteristic data of the elderly person, including but not limited to movement trajectory, movement speed, dwell time, activity range, and device interaction behavior. The mental state sub-database stores mental state data of the elderly person after standardization of format, including but not limited to operation time, question and answer results, game interaction scores, voice emotion data, social interaction scores, and intermediate values ​​of happiness index calculation. The log sub-database stores the operation logs and data traceability information of the entire process, including but not limited to data acquisition logs (device acquisition time, device status, data transmission status), data preprocessing logs (processing time of each step, amount of data processed, and abnormal data marking), model call logs (time of data input to the model and model output results), and early warning handling logs (early warning trigger time, handling results, and feedback information).

[0030] Among them, such as Figure 4 As shown, the model building and optimization module 30 includes a model building unit 31, a model training unit 32, and a model optimization unit 33. The model building unit 31 combines the elderly’s historical health data and individual physiological characteristics (such as age, type of chronic disease, and basic health status) to build an initial anomaly prediction model based on LSTM (Long Short-Term Memory Network) to capture the temporal change patterns of physiological and behavioral data. The model training unit 32 uses preprocessed historical data from the elderly’s exclusive database as the training set and labels the data that deviate from the elderly’s personalized normal baseline to supervise the training of the initial anomaly prediction model. During the training process, the gradient descent method is used to adjust the model parameters. The model optimization unit 33 periodically collects new data and early warning response feedback data to iteratively optimize the trained anomaly prediction model. The anomaly prediction model comprises an input layer, an embedding layer, an LSTM feature extraction layer, a fully connected fusion layer, and an output layer, connected sequentially. The input layer receives multi-dimensional time-series data output from the data preprocessing module 20. The input consists of time-series feature vectors containing physiological (heart rate, blood pressure, blood oxygen, etc.) and behavioral dimensions (movement speed, dwell time, device interaction frequency, etc.). The embedding layer vectorizes the discrete behavioral features (e.g., elderly activity status: home / outdoor / rest, device interaction type: game / psychological questioning) and concatenates them with continuous physiological features to form a feature matrix of uniform dimension. The LSTM feature extraction layer... The feature extraction layer consists of multiple stacked LSTM units, each containing 128 hidden nodes. It captures the long-short-term dependencies of the temporal data through gating mechanisms such as forget gates, input gates, and output gates, outputting a temporal feature map of the corresponding dimension. The fully connected fusion layer comprises two fully connected layers. The first fully connected layer flattens the temporal feature map into a one-dimensional feature vector and performs a non-linear transformation using the ReLU activation function. The second fully connected layer performs dimensionality reduction and fusion on the transformed features, outputting a predicted feature vector. The output layer uses a linear activation function to map the fused predicted feature vector to the future... Predicted health data values ​​at any given time; The gradient descent method is the Adam adaptive gradient descent method, specifically expressed as: ; ; ; ; ; in, For the first First-order moment estimation This represents the current iteration number. The first-order moment attenuation coefficient, For the first First-order moment estimation For second-order moment estimation, The second-order moment attenuation coefficient, For the first Second-order moment estimation, For the loss function with respect to model parameters gradient, and This is the moment estimate after bias correction. For the updated model parameters, For the first Model parameters for the next iteration For learning rate, To prevent the minimum value where the denominator is 0; The loss function is a weighted fusion loss function of mean squared error and mean absolute error, used to measure the deviation between the model's predicted values ​​and the true values. Its expression is: ; in, For loss function, These are weighting coefficients. The mean square error and satisfying , The mean absolute error and satisfying , The number of training samples, For the first The true value of each sample For the first The model prediction value for each sample.

[0031] Among them, such as Figure 5 As shown, the anomaly identification and early warning module 40 specifically includes: The real-time prediction unit 41 uses an anomaly prediction model to predict the physiological and behavioral data collected in real time by the data acquisition module 10 and outputs the predicted data for future moments. The deviation calculation unit 42 is used to calculate the relative deviation between the real-time acquired data, the model predicted data, and the individualized normal baseline of the elderly. The formula for calculating the relative deviation is: ,in, This refers to the relative deviation. for Real-time data values ​​collected at any given moment. for The corresponding personalized normal baseline value for the elderly at any given time, in the future The time interval can be 5 minutes, 15 minutes or 30 minutes, depending on the prediction step size setting during model training. The personalized normal baseline for the elderly is not a fixed value. The personalized normal baseline value for the elderly is updated by the model building and optimization module 30 based on the mean, variance and trend characteristics of the elderly's historical health data. For example, the blood pressure baseline of the elderly at 6 am will be generated by combining the average blood pressure in the morning within 3 months, the seasonal change trend and the characteristics of hypertension and chronic disease. The abnormality grading unit 43 is used to classify the abnormality level into slight deviation, moderate deviation and severe deviation according to the relative deviation. It does not judge an abnormality based on a single deviation, but monitors the relative deviation of multiple consecutive collection cycles. If the deviation exceeds the threshold for more than 3 consecutive cycles, it enters the abnormality grading stage to reduce the false alarm rate caused by temporary physiological fluctuations. The early warning push unit 44 pushes early warning information to the corresponding terminals according to the level of abnormality. When there is a slight deviation, it is pushed to the service personnel terminal 51; when there is a moderate deviation, it is pushed to the service personnel terminal 51, the family member terminal 53, and the resident terminal 54; when there is a severe deviation, it is pushed to the service personnel terminal 51, the family member terminal 53, and the community emergency equipment. The early warning information includes the elderly person's identity information, abnormal data, relative deviation degree, occurrence time, location information, potential risks, and handling suggestions.

[0032] Among them, such as Figure 6 As shown, the smart service interaction module 50 includes a service personnel terminal 51, an elderly terminal 52, a family member terminal 53, and a resident terminal 54. The service personnel terminal 51 is used to receive early warning information, view abnormal data and analysis reports, implement targeted intervention measures, and provide feedback on the handling results. The service personnel terminal 51 is a professional elderly care service workstation, such as a mobile APP, tablet, or computer, aimed at professional entities such as community elderly care workers, elderly care service center administrators, and medical personnel. The elderly terminal 52 adopts an age-friendly interface, supports viewing its own real-time health data, receiving simple early warning reminders, and initiating emergency calls. The elderly terminal 52 is an age-friendly dedicated terminal. Examples of terminals include age-friendly smart bracelets and tablets designed for the elderly, targeting the elderly population as the recipients of elderly care services. Family terminal 53 is used to view the elderly's real-time data, abnormal alerts, intervention reports, and mutual assistance records, and to communicate in real-time with service personnel terminal 51. Family terminal 53 is a remote monitoring terminal, such as a mobile APP, targeting the elderly's children, relatives, and other guardians. Resident terminal 54 is used to receive mutual assistance reminders, view authorized elderly data, and initiate mutual assistance responses, achieving collaborative care among community residents. Resident terminal 54 is a community mutual assistance terminal, such as a mobile APP or mobile mini-program, targeting residents in the community who participate in elderly care mutual assistance.

[0033] Among them, such as Figure 7 As shown, the mindset management module 60 specifically includes: The happiness index calculation unit 61, based on the collected mental state data, uses a weighted summation formula to calculate the elderly happiness index, transforming the collected unstructured / semi-structured mental state data into a quantifiable elderly happiness index, specifically expressed as follows: ,in, The happiness index for the elderly ranges from 0 to 100. A higher value indicates a greater degree of happiness and better mental health among the elderly. The game interaction score (0-100) is calculated based on the interaction data of the elderly and age-friendly terminals in memory-based and puzzle games, including dimensions such as game completion rate, reasonable operation time, and level completion accuracy. The psychological questions were scored (0-100), based on the elderly's responses to a standardized well-being assessment scale, which covers core issues such as life satisfaction, emotional state, and social willingness. The voice emotion score (0-100) is determined by voice recognition and analysis technology, which assesses the emotion of the elderly person's voice data during interaction. The higher the proportion of positive emotions (pleasure, peace), the higher the score. A social interaction score (0-100) is calculated based on the elderly person's behavioral data (such as the frequency of interaction with family / neighbors, the number of times they participate in community activities, and their willingness to communicate proactively). , , and All are weighting coefficients for the corresponding terms, and satisfy the following conditions: The weighting coefficients can be adjusted based on the individual characteristics of the elderly (such as age, education level, and social skills). For example, the weights can be increased for elderly people who live alone or have low social frequency. (Voice emotion) and weighting (Psychological questioning) can be given higher weight for elderly people with weaker cognitive abilities. (Game interaction); The abnormal mindset identification unit 62 has a built-in preset happiness index threshold (e.g., the preset happiness index threshold is 60, and the threshold setting needs to be combined with the personalized baseline of the elderly's historical mindset data). When the elderly's happiness index does not meet the preset happiness index threshold, it is judged as an abnormal mindset, triggering a preliminary abnormal mindset judgment. It is combined with the proportion of negative emotions in the voice emotion data (e.g., the proportion of voice segments with negative emotions such as anxiety, depression, and irritability to the total interactive voice) for auxiliary judgment. For example: if the proportion of negative emotions is ≥30% (can be customized), then an abnormal mindset is confirmed. If the proportion of negative emotions is <30%, it is judged as a temporary emotional fluctuation and is not marked as an abnormal mindset. The judgment result is pushed to the service personnel terminal 51 and the family member terminal 53 to continuously track the elderly's happiness index and generate a daily / weekly / monthly trend chart of mindset changes. If it is found that the happiness index continues to decline, it will prompt that a targeted psychological intervention plan needs to be developed.

[0034] Among them, such as Figure 8 As shown, the resident mutual assistance module 70 specifically includes: The data authorization and sharing unit 71 is used to set data sharing permissions, allowing the elderly to share some non-sensitive physiological data to the designated community resident terminal 54. The elderly or their authorized family members can set data sharing permissions through the age-friendly terminal and family member terminal 53. It supports specifying sharing targets such as community neighbors, volunteers, building leaders, etc., and limits the scope of shared data, only opening non-sensitive physiological data such as heart rate and steps, as well as basic behavioral data. The mutual aid reminder unit 72 is used to push mutual aid reminders to authorized resident terminals 54 when the elderly person shows slight or moderate deviations, reminding residents to pay attention to the elderly person's condition. The mutual aid reminder is only triggered when the abnormality identification and early warning module 40 determines that the elderly person's health status shows slight or moderate deviations. If it is a severe deviation, it will be pushed to the community emergency department first, and only serve as an auxiliary reminder. The mutual aid reminder includes the elderly person's basic identity information, the type of abnormality (such as slightly high heart rate, sudden decrease in activity level), the time of the abnormality and the elderly person's current location, and simple attention suggestions (such as "suggest visiting the elderly person's home to check on their condition" and "remind the elderly person to take their medication on time"). Mutual aid record unit 73 records the initiation, response, and completion of residents' mutual aid, forming a mutual aid log, which is stored in the elderly's exclusive database to record residents' mutual aid behavior throughout their entire life cycle, achieving centralized management and querying.

[0035] Example 2 The difference from Example 1 is that, as in Example 1, Figure 9 As shown, this invention provides a smart elderly care service management method with an anomaly alert function, applied to the smart elderly care service management system with an anomaly alert function shown in Embodiment 1 above. The method provided in this embodiment complements the system in Embodiment 1, jointly realizing smart elderly care services. The method specifically includes the following steps: Step S1: Collect the elderly’s physiological data, behavioral data and mental data in real time through several wearable monitoring devices and mental data collection devices 12, and transmit them to the data preprocessing module 20. Step S2: The data preprocessing module 20 preprocesses the collected raw data, classifies and stores the preprocessed data into the corresponding sub-databases, and builds a database specifically for the elderly. Step S3: Based on historical data in the database specifically for the elderly, the model building and optimization module 30 builds an initial anomaly prediction model based on LSTM. Through supervised training, cross-validation, and iterative optimization, an anomaly prediction model that meets the prediction accuracy requirements is obtained. Step S4: Input the real-time collected data into the anomaly prediction model for prediction. The relative deviation between the real-time collected data, the model prediction data, and the elderly person's personalized normal baseline is used to classify the anomalies based on the magnitude of the relative deviation. Step S5: Based on the anomaly classification, push the early warning information to the corresponding terminal through the smart service interaction module 50, and generate an anomaly analysis report and targeted intervention measures. Service personnel and family members perform the handling operations based on the early warning information and targeted intervention measures and provide feedback on the results. In this embodiment, the smart elderly care service management method further includes: Based on the collected mental state data, the happiness index of the elderly is calculated and compared with the preset happiness index threshold to identify abnormal mental states and trigger reminders. The elderly and their families can set data sharing permissions through the resident mutual assistance module 70 to share non-sensitive data to the designated resident terminal 54. When the elderly experience abnormalities, a mutual assistance reminder will be pushed to the authorized resident terminal 54 and the mutual assistance process will be recorded. Regularly summarize the collected data and mutual assistance records, and feed them back to the model building and optimization module 30 to iteratively optimize the anomaly prediction model and form a closed-loop management.

[0036] To further demonstrate the application value of the smart elderly care service management system and method with anomaly alert function in this invention, specific data is provided as follows.

[0037] Scenario 1: Application in community-based home care for the elderly. Taking Mr. Zhang, a 65-year-old elderly man living alone in a community (with a 5-year history of hypertension), as an example, the specific application process of the system of this invention is illustrated: We equipped Mr. Zhang with a smart bracelet with built-in physiological sensors and positioning units (collecting data every 15 minutes), and configured an age-friendly voice interaction terminal. The system collects his physiological data such as heart rate, blood pressure, and blood oxygen in real time, as well as behavioral data such as movement trajectory and activity range. It also collects his mental state data through games and psychological questions. The data preprocessing module 20 performs median filtering (to remove noise from blood pressure data), linear interpolation (to fill in missing data due to brief Bluetooth interruptions), and min-max normalization on the collected raw data. The processed data is then stored in Mr. Zhang's personalized database, with his personalized blood pressure baseline set to 135 / 85 mmHg. The model building and optimization module 30 constructs an LSTM anomaly prediction model based on Mr. Zhang's one month of historical health data and his history of hypertension. The model is trained using the Adam algorithm and a weighted fusion loss function to obtain the final anomaly prediction model. One morning, a smart bracelet recorded Mr. Zhang's blood pressure at 160 / 95 mmHg. After inputting the data into a model, it predicted that his blood pressure would rise to 165 / 98 mmHg in 30 minutes. The relative deviation was then calculated. Based on the relative deviation threshold (set according to actual needs), it is determined to be a moderate deviation; The smart service interaction module 50 pushes the warning information to the community elderly care specialist terminal and Mr. Zhang's children's terminal. The warning information includes "Mr. Zhang (Unit 3, Building 1, XX Community), blood pressure 160 / 95 mmHg, deviation 18.5%, moderate deviation, potential risk of hypertension attack, it is recommended to visit him at home immediately to measure blood pressure and guide him to take medication". If the children are not at home, the elderly care specialist will visit Mr. Zhang to measure his blood pressure and guide him to take antihypertensive drugs. The children can learn about the handling process in real time through the family terminal 53. The elderly care specialist will feed back the handling results to the system. The system collects Mr. Zhang's mental state data daily and calculates his happiness index to be 75 points (above the threshold of 60 points), indicating a normal mental state. Mr. Zhang and his children share non-sensitive data such as his steps and sleep quality with their neighbor, Ms. Li's, terminal. One day, Mr. Zhang's steps decreased by 80% compared to usual, and the system pushed a mutual assistance reminder to Ms. Li's terminal. Ms. Li visited him and found that Mr. Zhang was experiencing leg discomfort. She promptly informed the elderly care specialist, thus forming a community mutual assistance mechanism.

[0038] Scenario 2: Large-scale application in community-based elderly care service centers. Taking a community-based elderly care service center managing 200 elderly people as an example, the large-scale application process of the method of this invention is illustrated: Wearable monitoring devices and mental health data collection devices were configured for 200 elderly people in batches. The collection frequency was set according to the elderly people's health status (30 minutes / time for healthy elderly people, 15 minutes / time for elderly people with chronic diseases, and 5 minutes / time for disabled elderly people). The system established a database dedicated to 200 elderly people to realize data collection and storage by category. The model building and optimization module 30 constructs a community-wide LSTM anomaly prediction model, which is trained based on historical data of 200 elderly people. At the same time, personalized parameter adjustments are made for each elderly person's individual characteristics (age, chronic disease history, physical condition). The real-time data of the 200 elderly people are monitored in batches and anomalies are identified. The model is graded into mild / moderate / severe deviations and issued early warnings. The community elderly care service center sets up 3 elderly care specialists to handle mild / moderate deviations, and the community hospital is responsible for emergency treatment of severe deviations, so as to achieve efficient division of labor in early warning and handling. The mindset management module monitors the happiness index of 200 elderly people in batches, provides psychological counseling to 15 elderly people whose happiness index is below the threshold, and establishes a community mutual aid team (50 community residents) based on the community residents' committee. The elderly and their families can choose mutual aid members independently, and the system realizes mutual aid reminders and records, forming a community collaborative care system.

[0039] In summary, the system provided by this invention, through the collaborative work of multiple modules, achieves real-time monitoring, accurate prediction, and tiered early warning of the health status of the elderly, effectively improving the level of intelligence in elderly care services.

[0040] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A smart elderly care service management system with anomaly alert function, characterized in that, include: The data acquisition module is used to collect physiological, behavioral, and psychological data of the elderly. The data preprocessing module is used to store and preprocess the collected data to build a database specifically for the elderly. The model building and optimization module trains and iteratively optimizes the model based on historical data in the database specifically for the elderly, resulting in the final anomaly prediction model. The anomaly identification and early warning module uses an anomaly prediction model to predict and compare real-time collected data, determine the degree of data deviation and trigger corresponding anomaly level warnings. The anomaly prediction model is a multi-dimensional time series prediction model based on LSTM, which is used to capture the time series change patterns and inter-dimensional correlation features of physiological and behavioral data, and realize the prediction of future moments and anomaly determination of elderly health data. The intelligent service interaction module is used to push early warning information, abnormal data, abnormal analysis reports and targeted intervention measures to service personnel, the elderly and their families; The mindset management module is used to analyze the mindset data of the elderly and monitor their happiness level. The resident mutual assistance module is used to enable authorized sharing of elderly people's physical data and mutual assistance reminders among residents; The modules communicate with each other through the network to achieve collaborative work.

2. The intelligent elderly care service management system with anomaly alert function according to claim 1, characterized in that, The data acquisition module includes several wearable monitoring devices and a mental health data acquisition device. The wearable monitoring devices are equipped with physiological sensors and positioning units to collect the elderly's dynamic physiological data and movement trajectories in real time. The physiological sensors include heart rate, blood pressure, blood oxygen, body temperature, and blood glucose sensors. The mental health data acquisition device uses an age-friendly interactive terminal to collect the elderly's mental health data by pushing games and psychological questions. The mini-games include memory-based and brain-training lightweight games. The psychological questions use a standardized happiness assessment scale and simultaneously collect the elderly's operation time, question and answer results, and voice emotion data during the interaction process.

3. The intelligent elderly care service management system with anomaly alert function according to claim 1, characterized in that, The data preprocessing module is used to filter, deduplicate, normalize, and fill missing values ​​in the raw data collected by the data acquisition module. Specifically, it uses median filtering to remove noise from physiological data, linear interpolation to fill in short-term missing data, and min-max normalization to perform normalization. The min-max normalization formula is expressed as: ,in, This is the original data. The minimum value of the data. The maximum value of the data is used to obtain preprocessed data after filtering, deduplication, normalization, and missing value imputation. The preprocessed data is used for model training and real-time analysis. The data preprocessing module has a built-in database specifically for the elderly, which is divided into physiological sub-database, behavioral sub-database, mental state sub-database and log sub-database according to data type.

4. The intelligent elderly care service management system with anomaly alert function according to claim 1, characterized in that, The model building and optimization module includes a model building unit, a model training unit, and a model optimization unit. The model building unit combines the elderly’s historical health data and individual physiological characteristics to build an initial anomaly prediction model based on LSTM, which is used to capture the temporal variation patterns of physiological and behavioral data. The model training unit uses preprocessed historical data from the elderly’s exclusive database as the training set and labels the data deviations from the elderly’s personalized normal baseline to supervise the training of the initial anomaly prediction model. During the training process, the gradient descent method is used to adjust the model parameters. The model optimization unit periodically collects new data and early warning response feedback data to iteratively optimize the trained anomaly prediction model. The anomaly prediction model comprises an input layer, an embedding layer, an LSTM feature extraction layer, a fully connected fusion layer, and an output layer connected in sequence. The input layer receives multi-dimensional time-series data output from the data preprocessing module, with the input being a time-series feature vector containing physiological and behavioral dimensions. The embedding layer vectorizes the discrete behavioral features of the input and concatenates them with continuous physiological features to form a feature matrix of a unified dimension. The LSTM feature extraction layer comprises multiple stacked LSTM units, capturing the long-short-term dependencies of the time-series data through a gating mechanism, and outputting a time-series feature map of the corresponding dimension. The fully connected fusion layer comprises two fully connected layers. The first fully connected layer flattens the time-series feature map into a one-dimensional feature vector and performs a non-linear transformation using the ReLU activation function. The second fully connected layer performs dimensionality reduction and fusion on the transformed features, outputting a predicted feature vector. The output layer uses a linear activation function to map the fused predicted feature vector to future... Predicted health data values ​​at any given time; The gradient descent method mentioned is the Adam adaptive gradient descent method, specifically expressed as follows: ； ； ； ； ; in, For the first First-order moment estimation This represents the current iteration number. The first-order moment attenuation coefficient, For the first First-order moment estimation For second-order moment estimation, The second-order moment attenuation coefficient, For the first Second-order moment estimation, For the loss function with respect to model parameters gradient, and This is the moment estimate after bias correction. For the updated model parameters, For the first Model parameters for the next iteration For learning rate, To prevent the minimum value where the denominator is 0; The loss function is a weighted fusion loss function of mean squared error and mean absolute error, used to measure the deviation between the model's predicted values ​​and the true values. Its expression is: ; in, For loss function, These are weighting coefficients. The mean square error and satisfying , The mean absolute error and satisfying , The number of training samples, For the first The true value of each sample For the first The model prediction value for each sample.

5. The intelligent elderly care service management system with anomaly alert function according to claim 1, characterized in that, The anomaly identification and early warning module specifically includes: The real-time prediction unit uses an anomaly prediction model to predict the physiological and behavioral data collected in real time by the data acquisition module and outputs the predicted data for future moments. The deviation calculation unit is used to calculate the relative deviation between the real-time collected data, the model predicted data, and the individualized normal baseline of the elderly. The formula for calculating the relative deviation is as follows: ,in, This refers to the relative deviation. for Real-time data values ​​collected at any given moment. for The personalized normal baseline value of the elderly at any given time is obtained by the model building and optimization module based on the mean, variance and trend characteristics of the elderly’s historical health data. Anomaly grading unit is used to classify anomalies into minor deviation, moderate deviation, and severe deviation based on their relative degree of deviation; The early warning push unit pushes early warning information to the corresponding terminals according to the level of abnormality. When there is a slight deviation, it is pushed to the service personnel terminal; when there is a moderate deviation, it is pushed to the service personnel terminal, family member terminal, and resident terminal; when there is a severe deviation, it is pushed to the service personnel terminal, family member terminal, and community emergency equipment. The early warning information includes the elderly person's identity information, abnormal data, relative deviation degree, occurrence time, location information, potential risks, and handling suggestions.

6. The intelligent elderly care service management system with anomaly alert function according to claim 1, characterized in that, The mindset management module specifically includes: The happiness index calculation unit, based on collected mental health data, uses a weighted summation formula to calculate the happiness index of the elderly, specifically expressed as follows: ,in, To improve the happiness index of the elderly. To score points for game interactions, Scoring based on psychological questions, To score the emotion of the voice, Scoring based on social interaction, , , and All are weighting coefficients for the corresponding terms, and satisfy the following conditions: ; The abnormal mindset recognition unit has a built-in preset happiness index threshold. When the elderly person's happiness index does not meet the preset happiness index threshold, it is judged as an abnormal mindset. The judgment is further aided by the proportion of negative emotions in the voice emotion data, and the judgment result is pushed to the service personnel terminal and the family member terminal.

7. The intelligent elderly care service management system with anomaly alert function according to claim 1, characterized in that, The resident mutual assistance module specifically includes: The data authorization and sharing unit is used to set data sharing permissions and share some non-sensitive physiological data of the elderly to designated community residents' terminals. The mutual assistance reminder unit is used to push mutual assistance reminders to authorized resident terminals when the elderly person shows slight or moderate deviation, reminding residents to pay attention to the elderly person's condition; The mutual aid record unit records the initiation, response, and completion of mutual aid among residents, forming a mutual aid log which is stored in a database specifically for the elderly.

8. A smart elderly care service management method with anomaly alert function, applied to the smart elderly care service management system with anomaly alert function according to any one of claims 1-7, characterized in that, Specifically, the following steps are included: Step S1: Collect the elderly’s physiological data, behavioral data and mental data in real time through several wearable monitoring devices and mental data collection devices, and transmit them to the data preprocessing module. Step S2: The collected raw data is preprocessed by the data preprocessing module, and the preprocessed data is classified and stored in the corresponding sub-databases to build a database specifically for the elderly. Step S3: Based on historical data in the database specifically for the elderly, the model building and optimization module builds an initial anomaly prediction model based on LSTM. Through supervised training, cross-validation, and iterative optimization, an anomaly prediction model that meets the prediction accuracy requirements is obtained. Step S4: Input the real-time collected data into the anomaly prediction model for prediction. The relative deviation between the real-time collected data, the model prediction data, and the elderly person's personalized normal baseline is used to classify the anomalies based on the magnitude of the relative deviation. Step S5: Based on the anomaly classification, push the early warning information to the corresponding terminal through the smart service interaction module, and generate an anomaly analysis report and targeted intervention measures. Service personnel and family members perform the handling operations based on the early warning information and targeted intervention measures and provide feedback on the results.

9. The smart elderly care service management method with anomaly alert function according to claim 8, characterized in that, The smart service interaction module includes service personnel terminals, elderly terminals, family member terminals, and resident terminals; The service personnel terminal is used to receive early warning information, view abnormal data and analysis reports, implement targeted intervention measures, and provide feedback on the handling results; The elderly terminal uses an age-friendly interface and supports viewing real-time health data, receiving simple early warning reminders, and initiating emergency calls. The family member terminal is used to view the elderly person's real-time data, abnormal warnings, intervention reports and mutual assistance records, and to communicate with the service personnel's terminal in real time. The resident terminal is used to receive mutual assistance reminders, view authorized elderly data, and initiate mutual assistance responses, thereby achieving collaborative care among community residents.

10. The smart elderly care service management method with anomaly alert function according to claim 8, characterized in that, The intelligent elderly care service management method also includes: Based on the collected mental state data, the happiness index of the elderly is calculated and compared with the preset happiness index threshold to identify abnormal mental states and trigger reminders. The elderly and their families can set data sharing permissions through the resident mutual assistance module, sharing non-sensitive data to designated resident terminals. When the elderly experience abnormalities, mutual assistance reminders are pushed to authorized resident terminals, and the mutual assistance process is recorded. Regularly summarize the collected data and mutual assistance records, and feed them back to the model building and optimization module to iteratively optimize the anomaly prediction model, forming a closed-loop management system.

Images