Human health management method based on intelligent sensor

Abstract

The invention discloses a human body health management method based on an intelligent sensor. The human body health management method comprises the following steps that original physiological signal data are collected through the intelligent sensor and preprocessed; mutual information maximization fusion is carried out to obtain fused physiological signal data; performing feature extraction to form a physiological signal feature matrix; performing health assessment by using the improved GAAE model to obtain a health assessment result; establishing a corresponding health threshold set, and generating a health management scheme; the health assessment result and the health management scheme are visually displayed on a user terminal, and a user health file is established; the improved GAAE model is incrementally updated, and the health management scheme is dynamically corrected, so that the high-precision recognition of the individual health state is realized, and the scientificity and continuity of human health management are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent health monitoring, and in particular to a method for human health management based on intelligent sensors. Background Technology

[0002] With the rapid development of smart wearable devices and biosensor technologies, human health monitoring is gradually shifting from traditional medical testing to a real-time, continuous, and personalized intelligent health management model. Existing health monitoring methods often rely on single physiological indicators for judgment, with limited data sources and a lack of comprehensive analytical capabilities, making it difficult to fully reflect an individual's health status. Especially in scenarios involving the simultaneous acquisition of multimodal physiological signals, current technologies often suffer from insufficient data fusion capabilities, poor robustness of analytical algorithms, and inadequate consideration of individual differences, resulting in insufficient accuracy in health assessment results and hindering personalized health management.

[0003] In traditional intelligent health management systems, physiological signal acquisition typically employs independent sensor devices, each operating independently and outputting raw signal data. Due to differences in the working mechanisms, sampling frequencies, and environmental noise levels of different sensors, time shifts and signal distortions often exist between data points, affecting the effectiveness of data fusion. Current research employs simple weighted averaging or feature stitching for multi-source data fusion; however, these methods do not consider the correlation and mutual information between signals, easily leading to information redundancy or loss of key features, thus impacting the stability and accuracy of health assessment models. Summary of the Invention

[0004] One objective of this invention is to propose a human health management method based on intelligent sensors. This invention makes full use of multimodal intelligent sensing, mutual information maximization fusion method, improved GAAE model and cloud health record incremental learning mechanism to realize real-time collection, intelligent assessment and dynamic management of human health status.

[0005] A human health management method based on intelligent sensors according to an embodiment of the present invention includes the following steps:

[0006] Raw physiological signal data is collected and preprocessed using intelligent sensors;

[0007] The preprocessed raw physiological signal data is fused by maximizing mutual information to obtain fused physiological signal data.

[0008] Feature extraction is performed based on fused physiological signal data to form a physiological signal feature matrix;

[0009] The physiological signal feature matrix is ​​input into the improved GAAE model for health assessment, and the health assessment results are obtained.

[0010] Based on the health assessment results, a corresponding set of health thresholds is established, and a health management plan is generated.

[0011] The health assessment results and health management plans are displayed visually on the user terminal, and the integrated physiological signal data, health assessment results and health management plans are uploaded to the cloud health management platform to establish user health records.

[0012] The system collects raw physiological signal data in real time, combines it with user health records to incrementally update the improved GAAE model, and dynamically corrects the health management plan to generate an updated health management plan.

[0013] Optionally, the raw physiological signal data includes heart rate, blood pressure, blood oxygen saturation, body temperature, movement posture, and sleep state, and the preprocessing includes signal filtering and noise reduction, abnormal data removal, and sampling time synchronization.

[0014] Optionally, obtaining the fused physiological signal data specifically includes:

[0015] The preprocessed raw physiological signal data is standardized to obtain standardized physiological signal data. The standardized physiological signal data is obtained by subtracting the mean from the sample of the preprocessed raw physiological signal data corresponding to each smart sensor and dividing by the standard deviation.

[0016] Time synchronization and alignment processing is performed on the standardized physiological signal data to obtain aligned standardized physiological signal data;

[0017] The aligned and standardized physiological signals are fused by maximizing mutual information to obtain fused physiological signal data. The fused physiological signal data is obtained by taking the aligned and standardized physiological signals from different sensors as multi-channel inputs, calculating the mutual information value between each signal channel, and the mutual information value refers to the information shared between signals. Based on the mutual information value, a weight is assigned to each channel, and the weights of channels with high information correlation are combined to obtain the data.

[0018] Optionally, the formation of the physiological signal feature matrix specifically includes:

[0019] Based on the fused physiological signal data, the time window is divided according to a fixed window length and a fixed window step size to form a set of fused physiological signal time segments arranged in chronological order;

[0020] Temporal features are extracted from each fused physiological signal time segment in the set of fused physiological signal time segments. Based on the temporal features, the mean, standard deviation, skewness, and kurtosis are calculated and arranged into a temporal feature vector in a predetermined order. The mean refers to the arithmetic mean of the values ​​of all time window samples. The standard deviation refers to the root mean square deviation of all time window samples from the mean. The skewness refers to the average of the third central moments of all time window samples after standardization. The kurtosis refers to the average of the fourth central moments of all time window samples after standardization.

[0021] A frequency domain transformation is performed on each fused physiological signal time segment in the set of fused physiological signal time segments to obtain a spectral representation. Based on the spectral representation, the main frequency, spectral centroid, and spectral bandwidth are extracted and arranged into a frequency domain feature vector in a predetermined order. The extraction process is to obtain the frequency with the largest amplitude in the spectral representation as the main frequency, calculate the average frequency of the amplitude weighted of all frequency components as the spectral centroid, and calculate the square root of the weighted variance of the frequency components relative to the spectral centroid as the spectral bandwidth.

[0022] The corresponding time-domain feature vector and frequency-domain feature vector are concatenated at the vector level to generate a comprehensive feature vector.

[0023] The comprehensive feature vectors are combined into a feature matrix in chronological order, and normalization is performed according to the feature dimension to form a physiological signal feature matrix.

[0024] Optionally, obtaining the health assessment results specifically includes:

[0025] The physiological signal feature matrix is ​​input into the improved GAAE model for health assessment. The improved GAAE model includes a graph attention encoder module, a decoding and reconstruction consistency evaluation module, and a health status discrimination and incremental update module. The graph attention encoder module refers to the introduction of a diffusion convolution-spectral filtering hybrid encoding mechanism to obtain a set of latent representation vectors. The decoding and reconstruction consistency evaluation module refers to the decoder generating a reconstructed signal sequence according to a time window and calculating the mean squared difference to obtain a window-level score vector. The health status discrimination and incremental update module refers to summarizing and discriminating according to a preset time weighting and threshold rule, and outputting the health assessment result.

[0026] In the graph attention encoder module, a hybrid coding mechanism of diffusion convolution-spectral filtering is introduced to aggregate and encode the physiological signal feature matrix, generating a set of latent representation vectors. The generation process involves constructing a random walk diffusion operator based on a preset reference matrix and performing K-order diffusion convolution to obtain a diffusion representation. A normalized Laplacian operator is used to apply K-order Chebyshev spectral filtering to the physiological signal feature matrix to obtain a spectral representation. A causal dilated one-dimensional convolution is applied in the time window dimension to obtain a temporal representation. The diffusion representation, the spectral representation, and the temporal representation are concatenated and linearly mapped in a predetermined order to obtain a set of latent representation vectors.

[0027] In the input decoding and reconstruction consistency evaluation module, a window-level scoring vector is generated based on the latent representation vector set. The generation process involves inputting the latent representation vector into the decoder, which then generates a reconstructed signal window by window. The difference between the reconstructed signal and the original input signal is calculated to obtain the mean squared difference as the reconstruction error of the window. The reconstructed signal is then arranged in chronological order and normalized to obtain the window-level scoring vector.

[0028] In the health status discrimination and incremental update module, the window-level score vector is weighted and summed in chronological order to obtain a weighted comprehensive score. The weighted comprehensive score is compared with a preset health threshold range, and the health level is determined based on the range in which the weighted comprehensive score is located, and the health assessment result is output.

[0029] Optionally, the generation of the health management plan specifically includes:

[0030] A set of corresponding health thresholds is established based on the health assessment results. The process of establishing the set of health thresholds involves individualizing the range of each health assessment value in the health assessment results.

[0031] The difference between the health assessment value and the health threshold of the health threshold set is used to form the indicator deviation result set;

[0032] Based on the deviation direction and magnitude of each health indicator in the deviation result set, calculate the health adjustment coefficient for each indicator. The health adjustment coefficient is equal to the absolute value of the deviation value of the health indicator divided by the absolute value of the corresponding health threshold. Summarize all the health adjustment coefficients in order of health indicators to form a health adjustment coefficient set.

[0033] The set of health adjustment coefficients and health assessment results are input into the health management rule base to generate a health management plan. The initial health management plan includes a dietary recommendation unit, an exercise prescription unit, and a sleep plan unit. The dietary recommendation unit is formed based on the user's basal metabolic rate, weight, and deviation results to obtain the total energy intake target and the ratio of the three macronutrients. The exercise prescription unit is formed based on the heart rate change trend and physical fitness assessment results to obtain the lower and upper limits of the individualized target exercise intensity range, exercise type, single duration, and weekly frequency. The sleep plan unit is formed based on the fluctuation characteristics of body temperature rhythm and sleep-related signals to obtain the sleep onset time, total sleep duration, and target distribution of each sleep stage.

[0034] Optionally, the establishment of the user health record specifically includes:

[0035] The health assessment results and health management plans are encapsulated according to a unified field specification and timestamps are added to form a visually displayed dataset.

[0036] The visualization interface module transmits the dataset to the user terminal, completes parameter mapping and graphic rendering based on the preset display template, and generates a visualization interface, which includes a health score, a time trend curve, and management suggestion items.

[0037] The data, which integrates physiological signal data, health assessment results and health management plans, are packaged into an upload data package with timestamps and the upload sequence number is recorded. The data packages are then uploaded to the cloud-based health management platform in sequence.

[0038] The cloud-based health management platform performs encrypted storage and indexing on each uploaded data packet to establish a user health profile. The establishment process involves storing the integrated physiological signal data, health assessment results, and health management plan into corresponding data structures, generating a time index based on the timestamp of the data packet, and creating a unique identifier for each user. The three types of data records are associated with the unique identifier, organized and archived in chronological order to form a user health profile. The user health profile includes integrated physiological signal data records, health assessment result records, and health management plan records.

[0039] Optionally, the generation of the updated health management plan specifically includes:

[0040] Real-time raw physiological signal data is collected through intelligent sensors to form a real-time raw physiological signal data set.

[0041] The real-time raw physiological signal data set is input into the improved GAAE model, and the model parameters are incrementally updated according to the small step size rule to obtain the updated improved GAAE model. The small step size rule refers to making minor corrections to the parameters of the improved GAAE model based on the error of the real-time raw physiological signal data.

[0042] New health assessment results are generated using the updated and improved GAAE model, and compared with historical health assessment results in the user's health record index by index to obtain a health change vector.

[0043] The health change vector is input into the health management plan adjustment process to dynamically modify the dietary recommendation unit, exercise prescription unit, and sleep plan unit in the health management plan, and generate an updated health management plan.

[0044] The beneficial effects of this invention are:

[0045] This invention optimizes the entire process of health management, from multi-source physiological signal acquisition, intelligent fusion, and feature extraction to health assessment and dynamic management, by introducing a mutual information maximization fusion method and an improved GAAE model. Compared with existing technologies, this invention demonstrates significant improvements and innovations in technical effectiveness. Through the multimodal mutual information fusion method, this invention effectively improves the information utilization rate between signals from different sensors, reduces data redundancy and signal interference, and makes the fused physiological signal data more stable and reliable in terms of temporal continuity and feature integrity. Combined with the improved GAAE model, this invention constructs a deep feature representation that integrates temporal domain features, frequency domain features, and channel structure correlation, achieving high-precision modeling of complex human physiological states and intelligent health risk assessment.

[0046] At the health management level, this invention, through an adaptive health management plan generation mechanism based on health assessment results, can dynamically adjust dietary recommendations, exercise prescriptions, and sleep plans in conjunction with the user's health records, thereby achieving truly personalized health intervention. Simultaneously, the cloud-based health management platform of this invention possesses secure data storage, indexing, and time-series update capabilities, continuously accumulating user health data to support online incremental learning of the model. Through progressive updates of model parameters, this invention achieves adaptive optimization of the system during long-term monitoring, significantly improving the timeliness and accuracy of health assessment results, and forming a closed-loop health management process of "collection—analysis—feedback—correction."

[0047] Therefore, this invention achieves the beneficial effects of improving the accuracy of multi-source physiological signal fusion, enhancing the adaptive capability of health assessment models, and optimizing the personalized generation effect of health plans. It enables intelligent, dynamic, and continuously optimized human health management while maintaining low power consumption and high response speed. This method not only overcomes the shortcomings of traditional health monitoring technologies, such as low fusion accuracy, lagging assessment model updates, and insufficient personalization, but also provides a highly reliable and intelligent technical implementation path for fields such as intelligent medicine, remote health monitoring, and sports rehabilitation management. Attached Figure Description

[0048] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0049] Figure 1 This is an overall flowchart of a human health management method based on intelligent sensors proposed in this invention;

[0050] Figure 2 This is a schematic diagram of the modular structure of an improved GAAE model for human health management based on intelligent sensors proposed in this invention. Detailed Implementation

[0051] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0052] refer to Figure 1-2 A human health management method based on intelligent sensors includes the following steps:

[0053] Raw physiological signal data is collected and preprocessed using intelligent sensors;

[0054] The preprocessed raw physiological signal data is fused by maximizing mutual information to obtain fused physiological signal data.

[0055] Feature extraction is performed based on fused physiological signal data to form a physiological signal feature matrix;

[0056] The physiological signal feature matrix is ​​input into the improved GAAE model for health assessment, and the health assessment results are obtained.

[0057] Based on the health assessment results, a corresponding set of health thresholds is established, and a health management plan is generated.

[0058] The health assessment results and health management plans are displayed visually on the user terminal, and the integrated physiological signal data, health assessment results and health management plans are uploaded to the cloud health management platform to establish user health records.

[0059] The system collects raw physiological signal data in real time, combines it with user health records to incrementally update the improved GAAE model, and dynamically corrects the health management plan to generate an updated health management plan.

[0060] In this embodiment, the raw physiological signal data includes heart rate, blood pressure, blood oxygen saturation, body temperature, exercise posture, and sleep state. The preprocessing includes signal filtering and noise reduction, abnormal data removal, and sampling time synchronization.

[0061] In this embodiment, obtaining the fused physiological signal data specifically includes:

[0062] The preprocessed raw physiological signal data is standardized to obtain standardized physiological signal data. The standardized physiological signal data is obtained by subtracting the mean from the sample of the preprocessed raw physiological signal data corresponding to each smart sensor and dividing by the standard deviation.

[0063] Time synchronization and alignment processing is performed on the standardized physiological signal data to obtain aligned standardized physiological signal data;

[0064] The aligned and standardized physiological signals are fused by maximizing mutual information to obtain fused physiological signal data. The fused physiological signal data is obtained by taking the aligned and standardized physiological signals from different sensors as multi-channel inputs, calculating the mutual information value between each signal channel, and the mutual information value refers to the information shared between signals. Based on the mutual information value, a weight is assigned to each channel, and the weights of channels with high information correlation are combined to obtain the data.

[0065] In this embodiment, the formation of the physiological signal feature matrix specifically includes:

[0066] Based on the fused physiological signal data, the time window is divided according to a fixed window length and a fixed window step size to form a set of fused physiological signal time segments arranged in chronological order;

[0067] Temporal features are extracted from each fused physiological signal time segment in the set of fused physiological signal time segments. Based on the temporal features, the mean, standard deviation, skewness, and kurtosis are calculated and arranged into a temporal feature vector in a predetermined order. The mean refers to the arithmetic mean of the values ​​of all time window samples. The standard deviation refers to the root mean square deviation of all time window samples from the mean. The skewness refers to the average of the third central moments of all time window samples after standardization. The kurtosis refers to the average of the fourth central moments of all time window samples after standardization.

[0068] A frequency domain transformation is performed on each fused physiological signal time segment in the set of fused physiological signal time segments to obtain a spectral representation. Based on the spectral representation, the main frequency, spectral centroid, and spectral bandwidth are extracted and arranged into a frequency domain feature vector in a predetermined order. The extraction process is to obtain the frequency with the largest amplitude in the spectral representation as the main frequency, calculate the average frequency of the amplitude weighted of all frequency components as the spectral centroid, and calculate the square root of the weighted variance of the frequency components relative to the spectral centroid as the spectral bandwidth.

[0069] The corresponding time-domain feature vector and frequency-domain feature vector are concatenated at the vector level to generate a comprehensive feature vector.

[0070] The comprehensive feature vectors are combined into a feature matrix in chronological order, and normalization is performed according to the feature dimension to form a physiological signal feature matrix.

[0071] In this embodiment, obtaining the health assessment results specifically includes:

[0072] The physiological signal feature matrix is ​​input into the improved GAAE model for health assessment. The improved GAAE model includes a graph attention encoder module, a decoding and reconstruction consistency evaluation module, and a health status discrimination and incremental update module. The graph attention encoder module refers to the introduction of a diffusion convolution-spectral filtering hybrid encoding mechanism to obtain a set of latent representation vectors. The decoding and reconstruction consistency evaluation module refers to the decoder generating a reconstructed signal sequence according to a time window and calculating the mean squared difference to obtain a window-level score vector. The health status discrimination and incremental update module refers to summarizing and discriminating according to a preset time weighting and threshold rule, and outputting the health assessment result.

[0073] In the graph attention encoder module, a hybrid coding mechanism of diffusion convolution-spectral filtering is introduced to aggregate and encode the physiological signal feature matrix, generating a set of latent representation vectors. The generation process involves constructing a random walk diffusion operator based on a preset reference matrix and performing K-order diffusion convolution to obtain a diffusion representation. A normalized Laplacian operator is used to apply K-order Chebyshev spectral filtering to the physiological signal feature matrix to obtain a spectral representation. A causal dilated one-dimensional convolution is applied in the time window dimension to obtain a temporal representation. The diffusion representation, the spectral representation, and the temporal representation are concatenated and linearly mapped in a predetermined order to obtain a set of latent representation vectors.

[0074] In the input decoding and reconstruction consistency evaluation module, a window-level scoring vector is generated based on the latent representation vector set. The generation process involves inputting the latent representation vector into the decoder, which then generates a reconstructed signal window by window. The difference between the reconstructed signal and the original input signal is calculated to obtain the mean squared difference as the reconstruction error of the window. The reconstructed signal is then arranged in chronological order and normalized to obtain the window-level scoring vector.

[0075] In the health status discrimination and incremental update module, the window-level score vector is weighted and summed in chronological order to obtain a weighted comprehensive score. The weighted comprehensive score is compared with a preset health threshold range, and the health level is determined based on the range in which the weighted comprehensive score is located, and the health assessment result is output.

[0076] In this embodiment, the generation of the health management plan specifically includes:

[0077] A set of corresponding health thresholds is established based on the health assessment results. The process of establishing the set of health thresholds involves individualizing the range of each health assessment value in the health assessment results.

[0078] The difference between the health assessment value and the health threshold of the health threshold set is used to form the indicator deviation result set;

[0079] Based on the deviation direction and magnitude of each health indicator in the deviation result set, calculate the health adjustment coefficient for each indicator. The health adjustment coefficient is equal to the absolute value of the deviation value of the health indicator divided by the absolute value of the corresponding health threshold. Summarize all the health adjustment coefficients in order of health indicators to form a health adjustment coefficient set.

[0080] The set of health adjustment coefficients and health assessment results are input into the health management rule base to generate a health management plan. The initial health management plan includes a dietary recommendation unit, an exercise prescription unit, and a sleep plan unit. The dietary recommendation unit is formed based on the user's basal metabolic rate, weight, and deviation results to obtain the total energy intake target and the ratio of the three macronutrients. The exercise prescription unit is formed based on the heart rate change trend and physical fitness assessment results to obtain the lower and upper limits of the individualized target exercise intensity range, exercise type, single duration, and weekly frequency. The sleep plan unit is formed based on the fluctuation characteristics of body temperature rhythm and sleep-related signals to obtain the sleep onset time, total sleep duration, and target distribution of each sleep stage.

[0081] In this embodiment, the establishment of the user's health record specifically includes:

[0082] The health assessment results and health management plans are encapsulated according to a unified field specification and timestamps are added to form a visually displayed dataset.

[0083] The visualization interface module transmits the dataset to the user terminal, completes parameter mapping and graphic rendering based on the preset display template, and generates a visualization interface, which includes a health score, a time trend curve, and management suggestion items.

[0084] The data, which integrates physiological signal data, health assessment results and health management plans, are packaged into an upload data package with timestamps and the upload sequence number is recorded. The data packages are then uploaded to the cloud-based health management platform in sequence.

[0085] The cloud-based health management platform performs encrypted storage and indexing on each uploaded data packet to establish a user health profile. The establishment process involves storing the integrated physiological signal data, health assessment results, and health management plan into corresponding data structures, generating a time index based on the timestamp of the data packet, and creating a unique identifier for each user. The three types of data records are associated with the unique identifier, organized and archived in chronological order to form a user health profile. The user health profile includes integrated physiological signal data records, health assessment result records, and health management plan records.

[0086] In this embodiment, the generation of the updated health management plan specifically includes:

[0087] Real-time raw physiological signal data is collected through intelligent sensors to form a real-time raw physiological signal data set.

[0088] The real-time raw physiological signal data set is input into the improved GAAE model, and the model parameters are incrementally updated according to the small step size rule to obtain the updated improved GAAE model. The small step size rule refers to making minor corrections to the parameters of the improved GAAE model based on the error of the real-time raw physiological signal data.

[0089] New health assessment results are generated using the updated and improved GAAE model, and compared with historical health assessment results in the user's health record index by index to obtain a health change vector.

[0090] The health change vector is input into the health management plan adjustment process to dynamically modify the dietary recommendation unit, exercise prescription unit, and sleep plan unit in the health management plan, and generate an updated health management plan.

[0091] Example 1:

[0092] This embodiment uses a smart wearable health monitoring scenario as an example to illustrate the application of the human health management method based on smart sensors proposed in this invention. The experiment selected 50 test subjects aged 25 to 60, including 20 office workers who sit for long periods, 20 people who engage in light exercise, and 10 people who engage in high-intensity exercise. Each subject wore a multimodal smart sensor device, with sensing modules including a heart rate sensor, blood pressure sensor, blood oxygen sensor, body temperature sensor, and triaxial accelerometer. The sampling frequency was set to 10 times per second, and the continuous monitoring period was 30 days.

[0093] In the experiment, raw physiological signal data were first collected using intelligent sensors and then filtered, denoised, and time-synchronized to form high-quality preprocessed data. Subsequently, a mutual information maximization fusion algorithm was used to fuse the multi-source signals, significantly improving the feature correlation between different sensor signals. Compared with the traditional weighted average fusion algorithm, mutual information fusion reduced data feature redundancy by approximately 24% and improved signal temporal consistency by approximately 15%.

[0094] During the data analysis phase, the system integrates physiological signal data input to improve the GAAE model for feature extraction and health assessment. This model introduces a joint mechanism of diffuse convolution and spectral filtering in graph structure encoding, effectively capturing the spatial dependence and temporal dynamics of multi-channel physiological signals. Compared to traditional autoencoder models and convolutional neural networks, the model of this invention reduces the reconstruction error of multimodal signals by an average of 17.8% and improves the accuracy of abnormal state identification by 11.6%.

[0095] During the health management phase, the system automatically generates individualized health management plans based on the assessment results, including dietary recommendations, exercise prescriptions, and sleep optimization plans. For example, for participants who are found to have a mildly elevated heart rate and insufficient sleep quality, the system calculates the lower limit of exercise intensity to be 110 bpm and the upper limit to be 138 bpm using a health adjustment factor, recommends 40 minutes of brisk walking daily, and suggests increasing the protein intake to 18% of total energy intake. This plan is visualized on the cloud-based health management platform and stored in the user's health record for subsequent follow-up.

[0096] On days 15 and 30 of system operation, health assessment results showed that most users experienced a significant improvement in their overall health scores. The average health score for office workers increased from an initial 68.3 to 81.2, sleep quality scores improved by approximately 14%, and the average blood oxygen saturation of exercisers remained stable above 98%. Compared to traditional static rule generation schemes, the method described in this invention demonstrates superior health management effectiveness in terms of intervention efficacy, personalized matching, and model stability.

[0097] To verify the effectiveness of the adaptive incremental learning mechanism of this invention, the experiment replaced the health record data of some users with new samples for online updates on the 20th day of monitoring. The results show that the improved GAAE model can maintain a health status recognition accuracy of over 93% without retraining, and the parameter convergence speed is improved by approximately 27% compared to the initial training phase. This demonstrates that the method possesses continuous learning and dynamic adaptation capabilities, enabling real-time health status adjustment and long-term individual tracking.

[0098] Table 1 Comparison of experimental data between the method of the present invention and traditional methods

[0099] User type Initial health score Health Score on Day 15 Health Score on Day 30 Average sleep quality improved (%) Blood oxygen stabilization rate (%) Model recognition accuracy (%) Health plan matching satisfaction (%) Office workers 68.3 76.5 81.2 14.0 97.2 93.5 91.6 People who exercise lightly 72.8 80.1 84.7 12.3 98.5 94.8 92.1 High-intensity exercise people 75.6 82.9 87.3 9.8 99.1 95.4 94.2 Control group (static rules) 70.4 72.6 74.1 5.7 95.8 86.2 78.5

[0100] As shown in Table 1, the human health management method based on intelligent sensors proposed in this invention significantly outperforms traditional methods in several key indicators. Through the mutual information maximization fusion algorithm, the stability and temporal consistency of the data are improved, making the input data of the health assessment model more representative and reliable. After 30 days of continuous testing, the experimental group showed an average improvement of 17.2% in health scores, while the control group only saw a 5.2% increase, indicating that the dynamic health management strategy of this invention can significantly improve individual health status.

[0101] Furthermore, in terms of sleep quality and blood oxygen stabilization rate, the experimental group showed an average improvement rate of 12.0% and 98.3%, respectively, while the control group only achieved 5.7% and 95.8%. This indicates that the improved GAAE model has higher sensitivity and accuracy in anomaly detection and physiological state recognition. User satisfaction survey results show that the personalized health management plan generated using this invention achieved an average satisfaction rate of 92.6%, significantly higher than the control group's 78.5%, demonstrating the significant advantages of this method in personalized intervention and user experience.

[0102] In summary, this invention establishes a complete closed-loop health management system through a collaborative mechanism of intelligent sensing and data acquisition, mutual information fusion, improved GAAE model, and cloud-based dynamic learning. This system not only improves the accuracy and personalization of health assessments but also possesses the ability for continuous optimization and real-time updates, making it widely applicable in scenarios such as chronic disease monitoring, exercise rehabilitation, and long-term health management.

Claims

1. A method for health management of a human body based on an intelligent sensor, characterized by, The method comprises the following steps: Collecting raw physiological signal data through intelligent sensors and preprocessing; Performing mutual information maximization fusion on the preprocessed raw physiological signal data to obtain fused physiological signal data; Extracting features based on the fused physiological signal data to form a physiological signal feature matrix; Inputting the physiological signal feature matrix into an improved GAAE model for health assessment to obtain a health assessment result; Based on the health assessment result, establishing a corresponding health threshold set to generate a health management scheme; Visualizing the health assessment result and the health management scheme on a user terminal and uploading the fused physiological signal data, the health assessment result and the health management scheme to a cloud health management platform to establish a user health record; Real-time collection of raw physiological signal data, incremental updating of the improved GAAE model in combination with the user health record, and dynamic correction of the health management scheme to generate an updated health management scheme.

2. The human health management method based on intelligent sensor according to claim 1, characterized in that, The raw physiological signal data includes heart rate, blood pressure, blood oxygen saturation, body temperature, exercise posture and sleep state, and the preprocessing includes signal filtering and denoising, abnormal data rejection and sampling time synchronization. 3.The human health management method based on the intelligent sensor according to claim 1, characterized in that, The fused physiological signal data is obtained specifically as follows: Standardizing the preprocessed raw physiological signal data to obtain standardized physiological signal data, which is obtained by subtracting the mean value and dividing by the standard deviation of the sample of the preprocessed raw physiological signal data corresponding to each intelligent sensor; Performing time synchronization and alignment processing on the standardized physiological signal data to obtain aligned standardized physiological signal data; Performing mutual information maximization fusion on the aligned standardized physiological signal to obtain the fused physiological signal data, which is obtained by taking the aligned standardized physiological signal of different sensors as multi-channel input, calculating the mutual information value between each signal channel, and combining the channel weights of high information correlation.

4. The human health management method based on the intelligent sensor according to claim 1, characterized in that, The formation of the physiological signal feature matrix specifically comprises: Based on the fused physiological signal data, performing time window division according to a fixed window length and a fixed window step to form a set of fused physiological signal time segments arranged in time sequence; Extracting time domain features from each fused physiological signal time segment in the set of fused physiological signal time segments, calculating the mean value, standard deviation, skewness and kurtosis based on the time domain features, and composing a time domain feature vector in a predetermined order, wherein the mean value is the arithmetic mean of all time window sample values, the standard deviation is the root mean square deviation of all time window samples relative to the mean value, the skewness is the average of the third central moment of all time window samples after standardization, and the kurtosis is the average of the fourth central moment of all time window samples after standardization. The frequency domain transformation is performed on each fusion physiological signal time segment in the fusion physiological signal time segment set to obtain a frequency spectrum representation, a main frequency, a spectrum centroid and a spectrum bandwidth are extracted based on the frequency spectrum representation, and a frequency domain feature vector is formed in a predetermined order, the extraction process is to obtain a frequency with the maximum amplitude in the frequency spectrum representation as the main frequency, to calculate an average frequency of all frequency component amplitudes as the spectrum centroid, and to calculate a weighted variance square root of the frequency components relative to the spectrum centroid as the spectrum bandwidth; The corresponding time domain feature vector and the frequency domain feature vector are spliced in a vector level to generate a comprehensive feature vector; The comprehensive feature vector is combined in a time sequence to form a feature matrix, and normalization processing is performed according to the feature dimension to form a physiological signal feature matrix.

5. The human health management method based on the intelligent sensor according to claim 1, characterized in that, The health assessment result specifically includes: The physiological signal feature matrix is input into an improved GAAE model for health assessment, the improved GAAE model includes a graph attention encoder module, a decoding and reconstruction consistency evaluation module, and a health state discrimination and incremental update module, the graph attention encoder module refers to introducing a diffusion convolution-spectral graph filtering hybrid encoding mechanism to obtain a set of latent representation vectors, the decoding and reconstruction consistency evaluation module refers to a decoder generating a reconstruction signal sequence according to a time window, and calculating an average square difference to obtain a window-level score vector, and the health state discrimination and incremental update module refers to summarizing and discriminating according to a preset time weighting and threshold rule to output a health assessment result; In the graph attention encoder module, a diffusion convolution-spectral graph filtering hybrid encoding mechanism is introduced to aggregate and encode the physiological signal feature matrix to generate a set of latent representation vectors, the generation process is to construct a random walk diffusion operator according to a preset reference matrix and perform K-order diffusion convolution to obtain a diffusion representation, to perform K-order Chebyshev spectral graph filtering on the physiological signal feature matrix through a normalized Laplacian operator to obtain a spectral representation, and to apply a causal empty one-dimensional convolution in a time window dimension to obtain a time sequence representation, to splice and linearly map the diffusion representation, the spectral representation and the time sequence representation in a predetermined order to obtain the set of latent representation vectors; In the input decoding and reconstruction consistency evaluation module, a window-level score vector is generated based on the set of latent representation vectors, the generation process is to input the latent representation vectors into a decoder, to generate a reconstruction signal by the decoder according to a time window, to calculate a difference between the reconstruction signal and the original input signal to obtain a reconstruction error in the form of an average square difference, and to arrange and normalize the window-level score vector in a time sequence to obtain the window-level score vector; In the health state discrimination and incremental update module, the window-level score vectors are weighted and summed in a time sequence to obtain a weighted comprehensive score, the weighted comprehensive score is compared with a preset health threshold interval, a health level is determined according to the interval where the weighted comprehensive score is located, and a health assessment result is output.

6. The human health management method based on the intelligent sensor according to claim 1, characterized in that, The health management scheme is generated specifically by: A corresponding health threshold set is established based on the health assessment result, the establishment process is to individually correct the range of each health assessment value in the health assessment result to obtain the health threshold set; An index deviation result set is formed by the difference between the health assessment value and the health threshold value of the health threshold set; A health adjustment coefficient of each index is calculated according to the deviation direction and the deviation amplitude of each health index in the index deviation result set, and the health adjustment coefficient is equal to the absolute value of the health index deviation value divided by the absolute value of the corresponding health threshold value. All health adjustment coefficients are collected in the order of health indexes to form a health adjustment coefficient set; The health adjustment coefficient set and the health assessment result are input into the health management rule library to generate a health management scheme. The initial health management scheme includes a diet suggestion unit, an exercise prescription unit and a sleep scheme unit. The diet suggestion unit is formed according to the user's basal metabolic rate, body weight and index deviation result set to obtain the total energy intake target and the proportion of three major nutrients. The exercise prescription unit is formed according to the heart rate change trend and the physical ability evaluation result to obtain the lower limit and upper limit of the individualized target exercise intensity interval, the exercise mode, the single duration and the weekly frequency. The sleep scheme unit is formed according to the body temperature rhythm and the fluctuation characteristics of the sleep-related signal to obtain the sleep time, the total sleep duration and the target distribution of each sleep stage.

7. The human health management method based on the intelligent sensor according to claim 1, characterized in that, The establishment of the user health record specifically includes: The health assessment result and the health management scheme are data-encapsulated according to a unified field specification and are attached with a time stamp to form a visual display data set; The visual display data set is transmitted to the visual interface module of the user terminal, parameter mapping and graphic rendering are completed according to a preset display template to generate a visual interface. The visual interface includes a health score, a time trend curve and a management suggestion item; The fusion physiological signal data, the health assessment result and the health management scheme are packaged into upload data packets according to the time stamp and are recorded with an upload serial number, and are sequentially uploaded to the cloud health management platform; In the cloud health management platform, each upload data packet is executed for encrypted storage and index labeling to establish a user health record. The establishment process is to store the fusion physiological signal data, the health assessment result and the health management scheme into corresponding data structures respectively, to generate a time index according to the time stamp of the data packet, and to create a unique identifier for each user. The three types of data records are associated with the unique identifier, organized and archived in time sequence to form a user health record. The user health record includes fusion physiological signal data records, health assessment result records and health management scheme records. 8.The human health management method based on the intelligent sensor according to claim 1, wherein, The generation of the updated health management scheme specifically includes: Real-time original physiological signal data are collected by intelligent sensors to form a real-time original physiological signal data set; The real-time original physiological signal data set is input into the improved GAAE model, and the model parameters are incrementally updated according to the small step rule to obtain an updated improved GAAE model. The small step rule refers to the micro-correction of the parameters of the improved GAAE model according to the error of the real-time original physiological signal data; A new health assessment result is generated by using the updated improved GAAE model, and is compared with the historical health assessment result in the user health record index by index to obtain a health change vector; The health change vector is input into a health management scheme adjustment process, and a diet suggestion unit, an exercise prescription unit and a sleep scheme unit in the health management scheme are dynamically corrected to generate an updated health management scheme.

Images