Community-assisted health management method and system for the elderly based on wearable devices

Abstract

The application belongs to the technical field of health management, and discloses a community mutual aid type old-age health management method and system based on a wearable device.The method comprises the following steps: a cloud data center is used to construct a data analysis model, an alarm generation model, a voice recognition model and an aid matching model, and is deployed on all edge computing base stations; a wearable device is used to collect real-time health monitoring data and real-time demand voice data of a user, and transmit the data to the edge computing base stations; the edge computing base stations are used to perform data analysis by using the data analysis model, and perform alarm generation by using the alarm generation model; the edge computing base stations are used to perform voice recognition by using the voice recognition model, and perform aid matching by using the aid matching model.The application solves the problems of low efficiency, high cost, low accuracy, poor practicability and low intelligentization degree in the prior art.

Description

Technical Field

[0001] This invention belongs to the field of health management technology, specifically relating to a community-based mutual-aid elderly health management method and system based on wearable devices. Background Technology

[0002] With the increasing aging of my country's population, health issues among the elderly are becoming increasingly prominent. The elderly commonly suffer from chronic diseases, declining self-care abilities, and loneliness, making their need for health management particularly urgent. Wearable device technology is constantly maturing, such as smart bracelets, smartwatches, and health monitoring devices, which can monitor the physiological indicators and exercise data of the elderly in real time, providing data support for elderly health management. Communities are the primary living spaces for the elderly; therefore, conducting mutual-aid elderly health management based in communities is beneficial for integrating community resources and improving the quality of life and well-being of the elderly.

[0003] Existing technologies for elderly health management have the following shortcomings:

[0004] 1) Most existing health management technologies rely on statistical methods and manual analysis, which are inefficient, require a high level of professional knowledge, are costly, and cannot detect potential anomalies in the health monitoring data of the elderly, resulting in low accuracy of health management.

[0005] 2) Existing health management technologies remain at the end-to-end single management model, lack an effective application ecosystem, and are not adapted to the actual living conditions of the elderly, resulting in poor practicality of health management.

[0006] 3) Although existing health management technologies have introduced sensor technology for data collection and transmission, this single function alone cannot meet the needs of health and safety management for the elderly, resulting in a low level of intelligence. Summary of the Invention

[0007] In order to address the problems of low efficiency, high cost, low accuracy, poor practicality, and low level of intelligence in existing technologies, the present invention aims to provide a community-based mutual assistance elderly health management method and system based on wearable devices.

[0008] The technical solution adopted in this invention is as follows:

[0009] A community-based mutual-aid elderly health management method based on wearable devices includes the following steps:

[0010] The cloud data center builds data analysis models, alarm generation models, speech recognition models, and assistance matching models, and deploys them on all edge computing base stations;

[0011] Wearable devices collect users' real-time health monitoring data and real-time voice data, and transmit them in encrypted form to edge computing base stations within the community communication range;

[0012] The edge computing base station uses a data analysis model to analyze real-time health monitoring data, and uses an alarm generation model to generate alarms based on the real-time health monitoring data and the obtained real-time data analysis results. The obtained real-time alarm signals are then encrypted and transmitted to several monitoring terminals within the community communication range.

[0013] The edge computing base station performs speech recognition using a speech recognition model based on real-time demand voice data. Based on the obtained real-time demand text data and real-time assistance information from several monitoring terminals, it performs assistance matching using an assistance matching model. Based on the obtained real-time assistance matching results, the real-time demand text data is encrypted and transmitted to the target monitoring terminal that has successfully matched the assistance.

[0014] Furthermore, in the cloud data center, data analysis models, alarm generation models, speech recognition models, and assistance matching models are built and deployed on all edge computing base stations, including the following steps:

[0015] The cloud data center collects historical health monitoring data and historical demand voice data from several users, and preprocesses them to obtain several preprocessed historical health monitoring data and several preprocessed historical demand voice data.

[0016] Data dimensionality reduction is performed on several preprocessed historical health monitoring data. Based on the obtained dimensionality-reduced historical health monitoring data, a data analysis model is constructed using a deep learning algorithm, and several historical data analysis results are generated.

[0017] Based on the analysis results of historical data, several dimensionality-reduced historical health monitoring data corresponding to anomalies are identified. A deep learning algorithm is used to construct an alarm generation model, generate several historical alarm signals, and set a corresponding preset signal broadcast script for each historical alarm signal.

[0018] Based on several preprocessed historical demand voice data, a speech recognition model is constructed using a speech signal processing algorithm, several historical demand text data are generated, and based on several historical demand text data and several historical assistance information, an assistance matching model is constructed using a swarm intelligence optimization algorithm.

[0019] Extract the model metadata of the data analysis model, alarm generation model, speech recognition model, and assistance matching model, and encrypt and broadcast the model metadata to all edge computing base stations that communicate with the cloud data center, so that the edge computing base stations can deploy the model based on the model metadata.

[0020] Furthermore, the data analysis model is built based on the LSTM-Elman algorithm;

[0021] The alarm generation model is built based on the M-CGN-MLP algorithm;

[0022] The speech recognition model is built based on the STFT-logfBank-AA-MLP-CRF algorithm;

[0023] The aid matching model is built based on the ICPO algorithm.

[0024] Furthermore, dimensionality reduction is performed on several preprocessed historical health monitoring data sets. Based on the obtained dimensionality-reduced historical health monitoring data, a data analysis model is constructed using a deep learning algorithm, and several historical data analysis results are generated, including the following steps:

[0025] Using the PCA method, we performed dimensionality reduction on several preprocessed historical health monitoring data to obtain several dimensionality-reduced historical health monitoring data and key feature sets.

[0026] Several dimensionality-reduced historical health monitoring data were divided into a pre-training set without real labels and a first training set and a first test set with real labels in a ratio of 2:6:2.

[0027] The LSTM-Elman algorithm is used to build an initial data analysis model, and the classifier of the initial data analysis model is pre-trained using a pre-training set to obtain a pre-trained data analysis model.

[0028] Using the first training set, the pre-trained data analysis model is optimized and trained to obtain an optimized data analysis model, and several historical data analysis results are generated.

[0029] The optimized data analysis model is tested using the first test set to obtain the first test accuracy. If the first test accuracy is greater than the first accuracy threshold, the final data analysis model is output; otherwise, optimization training continues.

[0030] Furthermore, based on the historical data analysis results, several dimensionality-reduced historical health monitoring data corresponding to anomalies are identified. A deep learning algorithm is used to construct an alarm generation model, generating several historical alarm signals. A corresponding preset signal broadcast script is then set for each historical alarm signal, including the following steps:

[0031] Based on the preset M anomaly monitoring indicators, the M-CGN-MLP algorithm is used to construct the initial alarm generation model. Several dimensionality-reduced historical health monitoring data with real labels and historical data analysis results showing anomalies are divided into the second training set and the second test set in a 7:3 ratio.

[0032] Using the second training set, the initial alarm generation model is optimized and trained to obtain an optimized alarm generation model, which generates several historical alarm signals and sets a corresponding preset signal broadcast script for each historical alarm signal.

[0033] The optimized alarm generation model is tested using the second test set to obtain the second test accuracy. If the second test accuracy is greater than the second accuracy threshold, the final alarm generation model is output; otherwise, optimization training continues.

[0034] Furthermore, based on several preprocessed historical demand speech data, a speech recognition model is constructed using a speech signal processing algorithm to generate several historical demand text data. Then, based on these historical demand text data and several historical assistance provision information, an assistance matching model is constructed using a swarm intelligence optimization algorithm, including the following steps:

[0035] The STFT-logfBank-AA-MLP-CRF algorithm was used to construct an initial speech recognition model, and several preprocessed historical speech data were divided into a third training set and a third test set in a 7:3 ratio.

[0036] Using the third training set, the initial speech recognition model is optimized and trained to obtain an optimized speech recognition model, and several historical requirement text data are generated.

[0037] The optimized speech recognition model is tested using the third test set to obtain the third test accuracy. If the third test accuracy is greater than the third accuracy threshold, the final speech recognition model is output; otherwise, the speech recognition model continues to be optimized and trained.

[0038] Based on information from several historical aid sources, an initial aid matching model was constructed using the ICPO algorithm, and several historical demand text data were divided into a fourth training set and a fourth test set in a 7:3 ratio.

[0039] Using the fourth training set, the initial aid matching model was optimized and trained to obtain an optimized aid matching model, and several historical demand text data were generated.

[0040] The optimized aid matching model is tested using the fourth test set to obtain the fourth test accuracy. If the fourth test accuracy is greater than the fourth accuracy threshold, the final aid matching model is output; otherwise, the optimization training of the aid matching model continues.

[0041] Furthermore, the edge computing base station performs data analysis using a data analysis model based on real-time health monitoring data. Based on the real-time health monitoring data and the obtained real-time data analysis results, it generates an alarm using an alarm generation model and transmits the resulting real-time alarm signal in encryption to several monitoring terminals within the community communication range. This includes the following steps:

[0042] Edge computing base stations use the PCA method to reduce the dimensionality of real-time health monitoring data based on key feature sets, resulting in dimensionality-reduced real-time health monitoring data.

[0043] Using a data analysis model, we extract real-time data features from the dimensionality-reduced real-time health monitoring data, and perform data analysis based on these features to obtain the real-time data analysis results.

[0044] If the real-time data analysis results indicate an anomaly, the dimensionality-reduced real-time health monitoring data will be input into the alarm generation model and the process will proceed to the next step; otherwise, health monitoring data collection will continue.

[0045] Using an alarm generation model, M real-time data encoding sequences of the dimensionality-reduced real-time health monitoring data are generated, and alarms are generated based on the M real-time data encoding sequences to obtain real-time alarm signals.

[0046] Based on the preset signal broadcast script corresponding to the real-time alarm signal, the real-time alarm signal is encrypted and transmitted to several corresponding monitoring terminals within the community communication range.

[0047] Furthermore, the edge computing base station performs speech recognition using a speech recognition model based on real-time demand voice data, performs assistance matching using an assistance matching model based on the obtained real-time demand text data and real-time assistance information provided by several monitoring terminals, and encrypts and transmits the real-time demand text data to the target monitoring terminal with successfully matched assistance based on the obtained real-time assistance matching results, including the following steps:

[0048] The edge computing base station uses a speech recognition model to perform STFT processing on real-time demand speech data to obtain a real-time spectrogram.

[0049] Real-time spectral features of real-time spectrograms and real-time acoustic features of real-time demand speech data are extracted, and the real-time spectral features and real-time acoustic features are fused to obtain real-time fused features.

[0050] Based on real-time fusion characteristics, speech recognition is performed to obtain real-time demand text data, and real-time assistance information is collected from several monitoring terminals within the community communication range.

[0051] Based on real-time demand text data and several real-time assistance provision information, an assistance matching model is used to perform assistance matching to obtain real-time assistance matching results.

[0052] Based on the target real-time assistance information provided by the target with the best real-time demand text data in the real-time assistance matching results, the real-time demand text data is encrypted and transmitted to the target monitoring terminal where assistance was successfully matched.

[0053] Furthermore, based on real-time demand text data and several real-time assistance provision information, an assistance matching model is used to perform assistance matching to obtain real-time assistance matching results, including the following steps:

[0054] Based on several real-time assistance information, the real-time search space boundary of the assistance matching model is set, and the combination sequence of real-time demand text data and real-time assistance information is encoded into the encoding vector of the assistance matching model according to the encoding format.

[0055] Based on the encoded vector, the solution is initialized in the boundary of the real-time search space to obtain several initial solutions. Based on these initial solutions, the auxiliary matching model is used to iteratively optimize and obtain the optimal solution.

[0056] Decode the optimal solution to obtain a real-time assistance matching result that includes real-time demand text data and optimal target real-time assistance information.

[0057] A community-based mutual aid elderly health management system based on wearable devices is used to realize a community-based mutual aid elderly health management method. The system includes a cloud data center and several edge computing base stations, several wearable devices and several monitoring terminals. The cloud data center is communicatively connected to several edge computing base stations in several communities. Each edge computing base station is communicatively connected to several wearable devices and several monitoring terminals within the communication range of the corresponding community.

[0058] The beneficial effects of this invention are as follows:

[0059] This invention discloses a community-based mutual-aid elderly health management method and system based on wearable devices. It introduces wearable devices, providing online health monitoring and voice input functions, improving the convenience and real-time nature of health management for the elderly. The constructed data analysis model can uncover potential anomalies in health monitoring data, improving efficiency and reducing costs. Upon detecting anomalies, an alarm generation model is used for accurate further analysis, dynamically adjusting alarm strategies based on the elderly's health status and needs to reduce false alarms and missed alarms. It can generate accurate alarm signals immediately in the event of an incident and alert community volunteers, social workers, relatives, and other users, improving the safety of the elderly and forming a mutually beneficial mutual-aid ecosystem. Considering the community environment where the elderly live and the actual situation of health management and response, a community-based mutual-aid elderly health management network is constructed through a cloud data center, several edge computing base stations, and several monitoring terminals, achieving effective application ecosystem construction and improving the practicality of health management. It realizes multiple functions, including real-time data acquisition based on wearable devices, rapid data analysis based on edge computing base stations, customized alarm generation and sending, and matching voice input needs with real-time assistance, greatly improving the intelligence level of health management for the elderly.

[0060] Other beneficial effects of the present invention will be further explained in the specific embodiments. Attached Figure Description

[0061] Figure 1 This is a flowchart of the community-based mutual-aid elderly health management method based on wearable devices in this invention.

[0062] Figure 2 This is a structural block diagram of the community-based mutual aid elderly health management system based on wearable devices in this invention. Detailed Implementation

[0063] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0064] Example 1:

[0065] like Figure 1 As shown, this embodiment provides a community-based mutual-aid elderly health management method based on wearable devices, including the following steps:

[0066] S1: Cloud data center, building data analysis models, alarm generation models, speech recognition models, and assistance matching models, and deploying them on all edge computing base stations, including the following steps:

[0067] S1-1: Cloud data center, collects historical health monitoring data and historical demand voice data of several users, and performs preprocessing to obtain several preprocessed historical health monitoring data and several preprocessed historical demand voice data.

[0068] Health monitoring data includes parameters such as heart rate, blood pressure, blood oxygen saturation, steps and activity level, sleep quality, heart rate variability, body temperature, fall detection, blood glucose level, electrocardiogram, and respiratory rate;

[0069] Preprocessing includes data cleaning, format conversion, and normalization. While eliminating duplicate and erroneous data, it performs standardization to eliminate magnitude differences and noise, and converts the data into a format that subsequent models can recognize. Preprocessing of speech data also includes noise reduction and windowing.

[0070] S1-2: Perform dimensionality reduction on several preprocessed historical health monitoring data sets. Based on the obtained dimensionality-reduced historical health monitoring data sets, use deep learning algorithms to construct a data analysis model and generate several historical data analysis results, including the following steps:

[0071] S1-1-1: Using Principal Component Analysis (PCA), the dimensionality of several preprocessed historical health monitoring data is reduced to obtain several dimensionality-reduced historical health monitoring data and key feature sets;

[0072] The high-dimensional historical health monitoring dataset is transformed into a low-dimensional dataset, and a set of potentially related variables is transformed into a set of linearly unrelated variables. At the same time, as much information in the original data is preserved, the amount of computation is reduced and the model training speed is improved.

[0073] S1-1-2: Divide several dimensionality-reduced historical health monitoring data into a pre-training set without real labels, and a first training set and a first test set with real labels, in a ratio of 2:6:2;

[0074] S1-1-3: Use the Long Short-Term Memory (LSTM) network-Elman algorithm to build an initial data analysis model, and use the pre-training set to pre-train the classifier of the initial data analysis model to obtain a pre-trained data analysis model;

[0075] The data analysis model is built on the LSTM-Elman algorithm and includes a data feature extraction module built on the LSTM algorithm and a classifier built on the Elman algorithm, which are connected in sequence.

[0076] The data feature extraction module is used to extract data features from health monitoring data, mine abnormal information in health monitoring data, and use a pre-trained classifier to perform data analysis based on real-time data features and output data analysis results. This enables the detection of anomalies in health monitoring data. Moreover, this model is a coarse anomaly detection model, which can quickly and accurately discover abnormal information in health monitoring data, improve the real-time performance of health monitoring, reduce the waste of computing resources, and proceed to the next step of fine anomaly detection after anomaly detection is detected.

[0077] S1-1-4: Using the first training set, optimize the pre-trained data analysis model to obtain an optimized data analysis model and generate several historical data analysis results;

[0078] S1-1-5: Use the first test set to test the optimized data analysis model and obtain the first test accuracy. If the first test accuracy is greater than the first accuracy threshold, output the final data analysis model; otherwise, continue optimization training.

[0079] S1-3: Based on the historical data analysis results, several dimensionality-reduced historical health monitoring data corresponding to anomalies are identified. A deep learning algorithm is used to construct an alarm generation model, generate several historical alarm signals, and set a corresponding preset signal broadcast script for each historical alarm signal, including the following steps:

[0080] S1-3-1: Based on the preset M anomaly monitoring indicators, the initial alarm generation model is constructed using the M-Conditional Generative Networks (CGN)-Multilayer Perceptron (MLP) algorithm. Several dimensionality-reduced historical health monitoring data with real labels and historical data analysis results indicating anomalies are divided into the second training set and the second test set in a 7:3 ratio.

[0081] The alarm generation model is built based on the M-CGN-MLP algorithm, and includes M data encoding modules built based on the CGN algorithm and alarm generation modules built based on the MLP algorithm. All M encoding modules are connected to the alarm generation module.

[0082] A Conditional Variational Autoencoder (CVAE) is constructed using multiple alarm generation models to independently detect M anomaly monitoring indicators, achieving precise anomaly identification in health monitoring data. The CVAE encodes the health monitoring data based on independent anomaly monitoring indicators, generating data encoding sequences. These indicators include abnormal movement, abnormal vital signs, and abnormal sleep, and the overall characteristic information of all parameters in the monitoring data is comprehensively analyzed. The alarm generation module generates alarms based on the M data encoding sequences generated by the CVAE, obtaining the corresponding alarm signals.

[0083] S1-3-2: Using the second training set, optimize and train the initial alarm generation model to obtain an optimized alarm generation model, generate several historical alarm signals, and set a corresponding preset signal broadcast script for each historical alarm signal;

[0084] S1-3-3: Use the second test set to test the optimized alarm generation model and obtain the second test accuracy. If the second test accuracy is greater than the second accuracy threshold, output the final alarm generation model; otherwise, continue optimization training.

[0085] S1-4: Based on several preprocessed historical demand speech data, a speech recognition model is constructed using a speech signal processing algorithm to generate several historical demand text data. Then, based on these historical demand text data and several historical assistance provision information, an assistance matching model is constructed using a swarm intelligence optimization algorithm, including the following steps:

[0086] S1-4-1: The initial speech recognition model is constructed using the Short-Time Fourier Transform (STFT)-logfBank-Acoustic Analysis (AA)-MLP-Conditional Random Field (CRF) algorithm, and several preprocessed historical demand speech data are divided into a third training set and a third test set in a 7:3 ratio;

[0087] The speech recognition model is built based on the STFT-logfBank-AA-MLP-CRF algorithm. The speech recognition model includes a speech processing module based on the STFT algorithm, a spectral feature extraction module based on the logfBank algorithm, an acoustic feature extraction module based on the AA algorithm, a feature fusion module based on the MLP algorithm, and a text annotation module based on the CRF algorithm. The speech processing module is connected to the spectral feature extraction module and the acoustic feature extraction module, respectively. The spectral feature extraction module and the acoustic feature extraction module are both connected to the feature fusion module, and the feature fusion module is connected to the text annotation module.

[0088] The STFT algorithm processes speech data using STFT to generate corresponding spectrograms. The logfBank algorithm is similar to the Mel-Frequency Cepstrum (MFCC) algorithm, both based on the fBank feature extraction results and then performing subsequent processing. The logfBank algorithm has a lower computational cost than the MFCC algorithm, and the features have higher correlation. The correlation between features can be better utilized by deep learning models to improve recognition accuracy and reduce computational cost. It is used to extract spectral features from the spectrogram. The AA algorithm has functions for calculating volume, pitch, formants, and duration to extract acoustic features from speech data. The MLP network fuses the spectral and acoustic features of the input speech at different time scales, improving the ability to represent key information. The encoder processes the above fused features to generate a context vector, which contains the semantic information of the input speech. CRF can capture the dependency between semantic information and text prediction labels, improving recognition accuracy and realizing the recognition of text sequence annotation.

[0089] The formula for the volume calculation function is:

[0090]

[0091] In the formula, E RMS Here, is the volume calculation function; n is the sampling point indicator; N is the total number of sampling points; x[n] is the amplitude of the required voice data x(n) at sampling point n;

[0092] The formula for the pitch calculation function is:

[0093] E PITCH =X[n]+ε

[0094] In the formula, E PITCH X[n] is the pitch calculation function, which generally uses an automatic pitch detection algorithm; X[n] is the automatic pitch detection function for the required speech data x(n) at sampling point n; ε is the pitch compensation value;

[0095] The formula for calculating the resonance peak function is:

[0096]

[0097] In the formula, For prediction data; α i For prediction coefficients; x(ni) is the required speech data; n is the sampling point indicator; i is the sampling point indicator, representing the sampling point i before the current sampling point n; p is the prediction order; E FFT The predicted data is after Fast Fourier Transform (FFT); FFT[*] is the Fast Fourier Transform function; f k W represents the resonant frequency; max[*] is the maximum value preservation function; k The resonant bandwidth is w[*], where w[*] is the bandwidth calculation function.

[0098] The formula for the duration calculation function is:

[0099]

[0100] In the formula, L is the duration calculation function; The duration of the audio measurement for sampling point n; β is the silence interval duration for sampling point n; β is the duration compensation value; n is the sampling point indication value; N is the total number of sampling points;

[0101] S1-4-2: Using the third training set, optimize and train the initial speech recognition model to obtain an optimized speech recognition model and generate some historical requirement text data.

[0102] S1-4-3: Use the third test set to test the optimized speech recognition model and obtain the third test accuracy. If the third test accuracy is greater than the third accuracy threshold, output the final speech recognition model; otherwise, continue to optimize and train the speech recognition model.

[0103] S1-4-4: Based on information from several historical aid sources, an initial aid matching model was constructed using the Improved Crested Porcupine Optimizer (ICPO) algorithm, and several historical demand text data were divided into a fourth training set and a fourth test set in a 7:3 ratio;

[0104] The aid matching model is built based on the ICPO algorithm and includes a vector encoding module, an iterative optimization module based on the ICPO algorithm, and a vector decoding module connected in sequence.

[0105] The vector encoding module encodes the combined sequence of demand text data and assistance provision information into an encoding vector for the assistance matching model according to the encoding format. The iterative optimization module initializes the solution within the corresponding search space boundary based on the encoding vector, obtaining several initial solutions. Based on these initial solutions, the assistance matching model is used to iteratively optimize and obtain the optimal solution. The vector decoding module decodes the optimal solution to obtain the assistance matching result. The assistance provision information refers to the assistance information that social workers, volunteers, and medical personnel at the monitoring terminal can provide.

[0106] The initial aid matching model was constructed using the ICPO algorithm, and several historical demand text data were divided into a fourth training set and a fourth test set in a 7:3 ratio, including the following steps:

[0107] S1-4-4-1: Based on information provided by several historical aids, define the historical search space boundary of the ICPO algorithm, take maximizing the matching degree value as the optimization objective of the ICPO algorithm, and set the fitness function of the ICPO algorithm according to the optimization objective;

[0108] S1-4-4-2: Combine historical demand text data and historical aid provision information into a sequence, define the encoding format of ICPO individuals in the ICPO algorithm, and set the IPCO population parameters and maximum number of iterations for the IPCO algorithm;

[0109] S1-4-4-3: Based on the historical search space boundary, encoding format, fitness function, IPCO population parameters, and maximum number of iterations, the initial aid matching model is constructed using the ICPO algorithm;

[0110] S1-4-4-4: The historical aid provision information that best matches the historical demand text data is used as the corresponding real label to obtain several historical demand text data with real labels. The several historical demand text data with real labels are divided into the fourth training set and the fourth test set in a 7:3 ratio.

[0111] S1-4-5: Using the fourth training set, optimize and train the initial aid matching model to obtain an optimized aid matching model and generate some historical demand text data.

[0112] S1-4-6: Use the fourth test set to test the optimized aid matching model and obtain the fourth test accuracy. If the fourth test accuracy is greater than the fourth accuracy threshold, output the final aid matching model; otherwise, continue to optimize and train the aid matching model.

[0113] S1-5: Extract the model metadata of the data analysis model, alarm generation model, speech recognition model and assistance matching model, and encrypt and broadcast the model metadata to all edge computing base stations that communicate with the cloud data center so that the edge computing base stations can deploy the model according to the model metadata;

[0114] S2: Wearable devices collect users' real-time health monitoring data and real-time voice data, and transmit them in encrypted form to edge computing base stations within the community communication range;

[0115] S3: An edge computing base station, based on real-time health monitoring data, performs data analysis using a data analysis model. Based on the real-time health monitoring data and the obtained real-time data analysis results, it generates an alarm using an alarm generation model and transmits the obtained real-time alarm signal in encryption to several monitoring terminals within the community communication range, including the following steps:

[0116] S3-1: Edge computing base station, based on the key feature set, uses the PCA method to perform data dimensionality reduction on real-time health monitoring data to obtain dimensionality-reduced real-time health monitoring data;

[0117] S3-2: The data feature extraction module of the data analysis model is used to extract the real-time data features of the dimensionality-reduced real-time health monitoring data, and a classifier is used to perform data analysis based on the real-time data features to obtain the real-time data analysis results.

[0118] S3-3: If the real-time data analysis results show an anomaly, input the dimensionality-reduced real-time health monitoring data into the alarm generation model and proceed to the next step; otherwise, continue collecting health monitoring data.

[0119] S3-4: Using the data encoding module of the alarm generation model, generate M real-time data encoding sequences of the dimensionality-reduced real-time health monitoring data, and use the alarm generation module to generate alarms based on the M real-time data encoding sequences to obtain real-time alarm signals;

[0120] S3-5: Based on the preset signal broadcast script corresponding to the real-time alarm signal, the real-time alarm signal is encrypted and transmitted to several corresponding monitoring terminals within the community communication range;

[0121] S4: The edge computing base station performs speech recognition using a speech recognition model based on real-time demand voice data. Based on the obtained real-time demand text data and real-time assistance information from several monitoring terminals, it performs assistance matching using an assistance matching model. Based on the obtained real-time assistance matching results, it encrypts and transmits the real-time demand text data to the target monitoring terminal that has successfully matched the assistance. This includes the following steps:

[0122] S4-1: Edge computing base station, which uses a speech processing module based on a speech recognition model to perform STFT processing on real-time demand speech data to obtain a real-time spectrogram;

[0123] S4-2: Use the spectrum feature extraction module to extract the real-time spectrum features of the real-time spectrogram, use the acoustic feature extraction module to extract the real-time acoustic features of the real-time demand speech data, and use the feature fusion module to fuse the real-time spectrum features and real-time acoustic features to obtain the real-time fused features.

[0124] The formula for a real-time spectrogram is:

[0125]

[0126] In the formula, X(t,f) is the real-time spectrogram; x pre (n') represents the real-time voice data required at time n'; w(n'-t) is the window function; n' and t are both time indicators; f is the frequency indicator;

[0127] Extracting real-time spectral features from a real-time spectrogram includes the following steps:

[0128] A-1: Obtain the Mel frequency cepstral coefficients corresponding to the real-time spectrogram;

[0129] A-2: Obtain the first and second differences corresponding to the Mel frequency cepstral coefficients;

[0130] The formula for obtaining the Mel frequency cepstral coefficients is:

[0131]

[0132] In the formula, MFCCs are the Mel frequency cepstral coefficients corresponding to the real-time spectrogram X(t,f); DCT[*] is the discrete cosine transform function; log(*) is the logarithmic function; |X(t,f)| is the modulus corresponding to the real-time spectrogram X(t,f); arg(X(t,f)) is the phase corresponding to the real-time spectrogram X(t,f); and j is a complex number.

[0133] The formulas for first-order and second-order differences are:

[0134]

[0135] In the formula, MFCCs are the Mel frequency cepstral coefficients corresponding to the real-time spectrogram X(t,f); d1 is the first-order difference; d2 is the second-order difference; and i' is the differential indication.

[0136] S4-3: Based on the real-time fusion characteristics, use the text annotation module to perform speech recognition, obtain real-time demand text data, and collect real-time assistance information from several monitoring terminals within the community communication range;

[0137] S4-4: Based on real-time demand text data and several real-time assistance provision information, use an assistance matching model to perform assistance matching and obtain real-time assistance matching results, including the following steps:

[0138] S4-4-1: Based on several real-time assistance provision information, set the real-time search space boundary of the assistance matching model, and encode the real-time demand text data and real-time assistance provision information combination sequence into the encoding vector of the assistance matching model according to the encoding format.

[0139] S4-4-2: Based on the encoded vector, initialize the solution within the real-time search space boundary to obtain several initial solutions. Then, based on these initial solutions, use the aid matching model to iteratively optimize and obtain the optimal solution. This includes the following steps:

[0140] S4-4-2-1: Based on the encoding vector of the ICPO individual and the preset ICPO population parameters, the Circle chaotic mapping sequence is used for initialization to generate several initial ICPO individuals of the initial ICPO population;

[0141] The formula is:

[0142]

[0143] In the formula, The initial ICPO individuals, i.e., the initial solutions, are the initial solutions for the Circle chaotic mapping; is the initial ICPO individual generated randomly; i' is the ICPO individual indicator;

[0144] S4-4-2-2: Introduce a cyclic population reduction mechanism to limit the number of individuals in the ICPO population parameters and obtain the updated ICPO population parameters for the next iteration;

[0145] The formula is:

[0146]

[0147] In the formula, S t+1 S represents the number of individuals in the ICPO population parameters during the (t+1)th iteration; t S represents the number of individuals in the ICPO population parameters during the t-th iteration; min is the minimum number of individuals in the ICPO population parameters; a' is the function evaluation parameter; V is the function evaluation cycle parameter; V max The loop parameters are used to evaluate the maximum function; t is an indicator of the number of iterations.

[0148] S4-4-2-3: Calculate the initial fitness value of the initial ICPO individuals in the initial ICPO population based on the fitness function;

[0149] fit = F(LL,LS)

[0150] In the formula, fit is the fitness function; F(*) is the matching degree calculation function; LL is the real-time demand text data; LS is the real-time assistance information provided.

[0151] S4-4-2-4: Based on the initial fitness value and the updated ICPO population parameters, use the first defense strategy, the second defense strategy, the third defense strategy and the fourth defense strategy to update the initial ICPO population and obtain the updated ICPO population.

[0152] The formula for the first defense strategy is:

[0153]

[0154] In the formula, For updated ICPO individuals within the first defense range; τ1 represents the initial ICPO individuals within the first defense range; τ2 represents a random number based on a normal distribution; and τ3 represents a random value in the interval [0,1]. This is the optimal solution within the first defensive range; The vector generated between the true optimal solution within the first defense range and the optimal solution randomly selected from the ICPO population; i' is the ICPO individual indicator; t is the iteration indicator;

[0155] The formula for the second defense strategy is:

[0156]

[0157] In the formula, For updated ICPO individuals within the second defense range; The initial ICPO individuals within the second defense range; τ3 is the upper limit vector for the search of the second defense range; τ3 is a random value in the interval [0,1]. These are the r1th and r2th initial ICPO individuals, respectively; r1 and r2 are two random integers between [1, S]. A vector generated between the true optimal solution within the second defense perimeter and the optimal solution randomly selected from the ICPO population;

[0158] The formula for the third defense strategy is:

[0159]

[0160] In the formula, For updated ICPO individuals within the third defense range; The initial ICPO individuals within the third defense range; This is the upper limit vector for the search of the third defense range; These are the r2 and r3 initial ICPO individuals, respectively; r3 is a random integer between [1, S]. The odor diffusion factor defined for the fitness function; λ t As a defensive factor; This is a parameter for controlling the search direction;

[0161] The formula for the fourth defense strategy is:

[0162]

[0163] In the formula, For updated ICPO individuals within the fourth defense range; The initial ICPO individuals within the fourth defense range; This represents the optimal solution within the fourth defense range; τ4 and τ5 are both random values ​​in the interval [0,1]; λ t As a defensive factor; This is a parameter for controlling the search direction; a' is the average force affecting the search direction; a' is the convergence rate factor.

[0164] S4-4-2-4: Use a dynamic reverse learning algorithm to perform dynamic reverse learning on the updated ICPO population to generate a dynamically reversed ICPO population;

[0165] The formula is:

[0166]

[0167] In the formula, For dynamically reversed ICPO individuals; γ' is the decreasing inertia coefficient; L max L min These are the maximum and minimum values ​​in the vector space, respectively. For updated ICPO individuals;

[0168] S4-4-2-5: Based on the fitness function, calculate the fitness value of all ICPO individuals in the updated ICPO population and the dynamically reversed ICPO population, and select the ICPO individual with the lowest fitness value as the optimal individual and retain the optimal individual.

[0169] S4-4-2-6: If the number of iterations for iterative optimization reaches the maximum number of iterations or the fitness value of the optimal individual meets the requirements, then the optimal solution corresponding to the optimal individual will be output.

[0170] S4-4-3: Decode the optimal solution to obtain the real-time assistance matching result, which includes real-time demand text data and the optimal target real-time assistance provision information;

[0171] S4-5: Based on the target real-time assistance information that is optimal in the real-time demand text data in the real-time assistance matching results, encrypt and transmit the real-time demand text data to the target monitoring terminal that has successfully matched the assistance.

[0172] Example 2:

[0173] like Figure 2 As shown, this embodiment provides a community-based mutual aid elderly health management system based on wearable devices to realize a community-based mutual aid elderly health management method. The system includes a cloud data center and several edge computing base stations, several wearable devices and several monitoring terminals. The cloud data center is communicatively connected to several edge computing base stations in several communities. Each edge computing base station is communicatively connected to several wearable devices and several monitoring terminals within the communication range of the corresponding community.

[0174] The edge computing base station is used to perform data analysis using a data analysis model based on real-time health monitoring data. Based on the real-time health monitoring data and the obtained real-time data analysis results, it uses an alarm generation model to generate alarms and transmits the obtained real-time alarm signals in encryption to several monitoring terminals within the community communication range. Based on real-time demand voice data, it uses a speech recognition model to perform speech recognition. Based on the obtained real-time demand text data and real-time assistance information provided by several monitoring terminals, it uses an assistance matching model to perform assistance matching and transmits the real-time demand text data in encryption to the target monitoring terminal that has successfully matched assistance based on the obtained real-time assistance matching results.

[0175] Wearable devices are used to collect users' real-time health monitoring data and real-time demand voice data, and transmit them in encrypted form to edge computing base stations within the community communication range;

[0176] The monitoring terminal is used to provide real-time assistance information and receive real-time alarm signals and real-time demand text data sent by the edge computing base station;

[0177] The cloud data center is used to build data analysis models, alarm generation models, speech recognition models, and assistance matching models, and is deployed on all edge computing base stations.

[0178] This invention discloses a community-based mutual-aid elderly health management method and system based on wearable devices. It introduces wearable devices, providing online health monitoring and voice input functions, improving the convenience and real-time nature of health management for the elderly. The constructed data analysis model can uncover potential anomalies in health monitoring data, improving efficiency and reducing costs. Upon detecting anomalies, an alarm generation model is used for accurate further analysis, dynamically adjusting alarm strategies based on the elderly's health status and needs to reduce false alarms and missed alarms. It can generate accurate alarm signals immediately in the event of an incident and alert community volunteers, social workers, relatives, and other users, improving the safety of the elderly and forming a mutually beneficial mutual-aid ecosystem. Considering the community environment where the elderly live and the actual situation of health management and response, a community-based mutual-aid elderly health management network is constructed through a cloud data center, several edge computing base stations, and several monitoring terminals, achieving effective application ecosystem construction and improving the practicality of health management. It realizes multiple functions, including real-time data acquisition based on wearable devices, rapid data analysis based on edge computing base stations, customized alarm generation and sending, and matching voice input needs with real-time assistance, greatly improving the intelligence level of health management for the elderly.

[0179] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A community-based mutual-aid elderly health management method based on wearable devices, characterized in that: Includes the following steps: The cloud data center builds data analysis models, alarm generation models, speech recognition models, and assistance matching models, and deploys them on all edge computing base stations; Wearable devices collect users' real-time health monitoring data and real-time voice data, and transmit them in encrypted form to edge computing base stations within the community communication range; The edge computing base station uses a data analysis model to analyze real-time health monitoring data, and uses an alarm generation model to generate alarms based on the real-time health monitoring data and the obtained real-time data analysis results. The obtained real-time alarm signals are then encrypted and transmitted to several monitoring terminals within the community communication range. The edge computing base station performs speech recognition using a speech recognition model based on real-time demand voice data. Based on the obtained real-time demand text data and real-time assistance information from several monitoring terminals, it performs assistance matching using an assistance matching model. Based on the obtained real-time assistance matching results, the real-time demand text data is encrypted and transmitted to the target monitoring terminal that has successfully matched the assistance.

2. The community-based mutual-aid elderly health management method based on wearable devices according to claim 1, characterized in that: In the cloud data center, data analysis models, alarm generation models, speech recognition models, and assistance matching models are built and deployed on all edge computing base stations, including the following steps: The cloud data center collects historical health monitoring data and historical demand voice data from several users, and preprocesses them to obtain several preprocessed historical health monitoring data and several preprocessed historical demand voice data. Data dimensionality reduction is performed on several preprocessed historical health monitoring data. Based on the obtained dimensionality-reduced historical health monitoring data, a data analysis model is constructed using a deep learning algorithm, and several historical data analysis results are generated. Based on the analysis results of historical data, several dimensionality-reduced historical health monitoring data corresponding to anomalies are identified. A deep learning algorithm is used to construct an alarm generation model, generate several historical alarm signals, and set a corresponding preset signal broadcast script for each historical alarm signal. Based on several preprocessed historical demand voice data, a speech recognition model is constructed using a speech signal processing algorithm, several historical demand text data are generated, and based on several historical demand text data and several historical assistance information, an assistance matching model is constructed using a swarm intelligence optimization algorithm. Extract the model metadata of the data analysis model, alarm generation model, speech recognition model, and assistance matching model, and encrypt and broadcast the model metadata to all edge computing base stations that communicate with the cloud data center, so that the edge computing base stations can deploy the model based on the model metadata.

3. The community-based mutual-aid elderly health management method based on wearable devices according to claim 2, characterized in that: The data analysis model described above is built based on the LSTM-Elman algorithm; The alarm generation model described above is constructed based on the M-CGN-MLP algorithm; The speech recognition model described above is constructed based on the STFT-logfBank-AA-MLP-CRF algorithm; The aid matching model described above is constructed based on the ICPO algorithm.

4. The community-based mutual-aid elderly health management method based on wearable devices according to claim 3, characterized in that: The dimensionality of several preprocessed historical health monitoring data sets is reduced. Based on the obtained dimensionality-reduced historical health monitoring data, a data analysis model is constructed using a deep learning algorithm, and several historical data analysis results are generated, including the following steps: Using the PCA method, we performed dimensionality reduction on several preprocessed historical health monitoring data to obtain several dimensionality-reduced historical health monitoring data and key feature sets. Several dimensionality-reduced historical health monitoring data were divided into a pre-training set without real labels and a first training set and a first test set with real labels in a ratio of 2:6:

2. The LSTM-Elman algorithm is used to build an initial data analysis model, and the classifier of the initial data analysis model is pre-trained using a pre-training set to obtain a pre-trained data analysis model. Using the first training set, the pre-trained data analysis model is optimized and trained to obtain an optimized data analysis model, and several historical data analysis results are generated. The optimized data analysis model is tested using the first test set to obtain the first test accuracy. If the first test accuracy is greater than the first accuracy threshold, the final data analysis model is output; otherwise, optimization training continues.

5. A community-based mutual-aid elderly health management method based on wearable devices according to claim 4, characterized in that: Based on historical data analysis results, several dimensionality-reduced historical health monitoring data points corresponding to anomalies are identified. A deep learning algorithm is used to construct an alarm generation model, generating several historical alarm signals. A corresponding preset signal broadcast script is then set for each historical alarm signal, including the following steps: Based on the preset M anomaly monitoring indicators, the M-CGN-MLP algorithm is used to construct the initial alarm generation model. Several dimensionality-reduced historical health monitoring data with real labels and historical data analysis results showing anomalies are divided into the second training set and the second test set in a 7:3 ratio. Using the second training set, the initial alarm generation model is optimized and trained to obtain an optimized alarm generation model, which generates several historical alarm signals and sets a corresponding preset signal broadcast script for each historical alarm signal. The optimized alarm generation model is tested using the second test set to obtain the second test accuracy. If the second test accuracy is greater than the second accuracy threshold, the final alarm generation model is output; otherwise, optimization training continues.

6. The community-based mutual-aid elderly health management method based on wearable devices according to claim 3, characterized in that: Based on several preprocessed historical demand speech data, a speech recognition model is constructed using speech signal processing algorithms to generate several historical demand text data. Then, based on these historical demand text data and several historical assistance provision information, an assistance matching model is constructed using a swarm intelligence optimization algorithm, including the following steps: The STFT-logfBank-AA-MLP-CRF algorithm was used to construct an initial speech recognition model, and several preprocessed historical speech data were divided into a third training set and a third test set in a 7:3 ratio. Using the third training set, the initial speech recognition model is optimized and trained to obtain an optimized speech recognition model, and several historical requirement text data are generated. The optimized speech recognition model is tested using the third test set to obtain the third test accuracy. If the third test accuracy is greater than the third accuracy threshold, the final speech recognition model is output; otherwise, the speech recognition model continues to be optimized and trained. Based on information from several historical aid sources, an initial aid matching model was constructed using the ICPO algorithm, and several historical demand text data were divided into a fourth training set and a fourth test set in a 7:3 ratio. Using the fourth training set, the initial aid matching model was optimized and trained to obtain an optimized aid matching model, and several historical demand text data were generated. The optimized aid matching model is tested using the fourth test set to obtain the fourth test accuracy. If the fourth test accuracy is greater than the fourth accuracy threshold, the final aid matching model is output; otherwise, the optimization training of the aid matching model continues.

7. A community-based mutual-aid elderly health management method based on wearable devices according to claim 5, characterized in that: An edge computing base station performs data analysis using a data analysis model based on real-time health monitoring data. Based on the real-time health monitoring data and the obtained real-time data analysis results, it generates an alarm using an alarm generation model and transmits the encrypted real-time alarm signal to several monitoring terminals within the community communication range. The process includes the following steps: Edge computing base stations use the PCA method to reduce the dimensionality of real-time health monitoring data based on key feature sets, resulting in dimensionality-reduced real-time health monitoring data. Using a data analysis model, we extract real-time data features from the dimensionality-reduced real-time health monitoring data, and perform data analysis based on these features to obtain the real-time data analysis results. If the real-time data analysis results indicate an anomaly, the dimensionality-reduced real-time health monitoring data will be input into the alarm generation model and the process will proceed to the next step; otherwise, health monitoring data collection will continue. Using an alarm generation model, M real-time data encoding sequences of the dimensionality-reduced real-time health monitoring data are generated, and alarms are generated based on the M real-time data encoding sequences to obtain real-time alarm signals. Based on the preset signal broadcast script corresponding to the real-time alarm signal, the real-time alarm signal is encrypted and transmitted to several corresponding monitoring terminals within the community communication range.

8. A community-based mutual-aid elderly health management method based on wearable devices according to claim 6, characterized in that: An edge computing base station performs speech recognition using a speech recognition model based on real-time demand voice data. Based on the obtained real-time demand text data and real-time assistance information from several monitoring terminals, it performs assistance matching using an assistance matching model. Based on the obtained real-time assistance matching results, the real-time demand text data is encrypted and transmitted to the target monitoring terminal with the successfully matched assistance. The process includes the following steps: The edge computing base station uses a speech recognition model to perform STFT processing on real-time demand speech data to obtain a real-time spectrogram. Real-time spectral features of real-time spectrograms and real-time acoustic features of real-time demand speech data are extracted, and the real-time spectral features and real-time acoustic features are fused to obtain real-time fused features. Based on real-time fusion characteristics, speech recognition is performed to obtain real-time demand text data, and real-time assistance information is collected from several monitoring terminals within the community communication range. Based on real-time demand text data and several real-time assistance provision information, an assistance matching model is used to perform assistance matching to obtain real-time assistance matching results. Based on the target real-time assistance information provided by the target with the best real-time demand text data in the real-time assistance matching results, the real-time demand text data is encrypted and transmitted to the target monitoring terminal where assistance was successfully matched.

9. A community-based mutual-aid elderly health management method based on wearable devices according to claim 8, characterized in that: Based on real-time demand text data and several real-time assistance provision information, an assistance matching model is used to perform assistance matching to obtain real-time assistance matching results, including the following steps: Based on several real-time assistance information, the real-time search space boundary of the assistance matching model is set, and the combination sequence of real-time demand text data and real-time assistance information is encoded into the encoding vector of the assistance matching model according to the encoding format. Based on the encoded vector, the solution is initialized in the boundary of the real-time search space to obtain several initial solutions. Based on these initial solutions, the auxiliary matching model is used to iteratively optimize and obtain the optimal solution. Decode the optimal solution to obtain a real-time assistance matching result that includes real-time demand text data and optimal target real-time assistance information.

10. A community-based mutual-aid elderly health management system based on wearable devices, used to implement the community-based mutual-aid elderly health management method as described in any one of claims 1-9, characterized in that: The system includes a cloud data center, several edge computing base stations, several wearable devices, and several monitoring terminals. The cloud data center is communicatively connected to several edge computing base stations in several communities. Each edge computing base station is communicatively connected to several wearable devices and several monitoring terminals within the communication range of the corresponding community.

Images