Intelligent watch health information monitoring and management system based on AI technology

Abstract

The invention discloses a smart watch health information monitoring and management system based on an AI technology, and relates to the technical field of smart watch health information monitoring, and the system comprises a data collection module which collects various physiological and environmental data; the data preprocessing module is used for processing data by using improved mean filtering, linear interpolation and normalization; the health state analysis module is used for constructing a model by using various neural networks, analyzing the health state according to the multi-data feature vector and predicting the disease risk; the health information management module is used for storing data, building archives, giving personalized suggestions to the user and being in butt joint with a third-party platform; and the interaction module comprises a plurality of interaction devices, can check information, set parameters and receive notices, and is additionally provided with units such as exercise data acquisition and model optimization to assist the realization of system functions. The multi-element high-precision sensor is integrated to accurately collect data, the AI algorithm accurately analyzes the health state and the disease risk, double storage guarantees data safety, sharing is facilitated, the multi-interaction mode is convenient to operate, and user health is comprehensively protected.

Description

Technical Field

[0001] The present invention relates to the technical field of smart watch health information monitoring, and in particular to a smart watch health information monitoring and management system based on AI technology. Background Art

[0002] As people's living standards improve and their health awareness increases, the demand for personal health monitoring and management is growing. Traditional health monitoring methods often rely on professional equipment in hospitals and regular physical examinations, which have limitations and cannot monitor personal health in real time and continuously.

[0003] In modern society, people lead fast-paced and stressful lives, and many chronic diseases are becoming younger. The incidence of cardiovascular diseases and mental illnesses is increasing year by year. In order to detect health problems in a timely manner and take intervention measures, wearable health monitoring devices have emerged. As a common wearable device, smart watches have become a popular choice in the field of health monitoring due to their portability and ease of use.

[0004] However, early smart watches had many shortcomings in health monitoring. In terms of data collection, the types of physiological data collected were limited, and the sensor accuracy was not high enough. For example, some watches can only simply measure heart rate, and the measurement of important physiological indicators such as blood pressure and blood oxygen saturation is not accurate enough. At the same time, the function of collecting environmental data is even more lacking, and it is impossible to comprehensively consider the impact of environmental factors on human health.

[0005] In terms of data processing, early smart watches lacked an effective data preprocessing mechanism. The collected data often contained a lot of noise and outliers. Direct analysis of these unprocessed data would lead to inaccurate health status judgments. In addition, traditional data processing methods are simple and cannot fully mine the effective information in the data.

[0006] There is no advanced AI algorithm support for health status analysis. Early devices usually use a simple threshold judgment method to evaluate health status, which is too rough to accurately identify sub-health status and early disease risks. For example, in terms of cardiovascular disease risk prediction, relying solely on a single heart rate indicator is far from enough. A comprehensive analysis of heart rate variability, blood pressure fluctuations, and environmental factors is required.

[0007] In terms of health information management, the storage and management of user health records are not perfect. Local storage capacity is limited, data is easily lost, and there is a lack of data sharing mechanism with third-party medical platforms, which makes it impossible for doctors to obtain users' complete health information in a timely manner, which is not conducive to remote diagnosis and treatment.

[0008] In terms of interaction, early smart watches had simple interaction functions and poor voice interaction functions, which could not meet the diverse operational needs of users. At the same time, in terms of system upgrades, firmware upgrades were inconvenient and unsafe, which could easily lead to data loss or system failures.

[0009] To sum up, there are many problems that need to be solved in the existing smart watches in terms of health information monitoring and management. The smart watch health information monitoring and management system based on AI technology proposed in this patent is precisely to overcome these shortcomings and provide users with more accurate, comprehensive and convenient health monitoring and management services. Summary of the invention

[0010] The present invention proposes a smart watch health information monitoring and management system based on AI technology to solve the problems mentioned in the above-mentioned prior art.

[0011] In order to achieve the above objectives, the present invention adopts the following technical solution: a smart watch health information monitoring and management system based on AI technology, comprising:

[0012] Data acquisition module: integrates multiple sensors to collect multiple physiological data and environmental data of users in real time;

[0013] Data preprocessing module: preprocessing the collected physiological data and environmental data, data cleaning, removing outliers and noise. The noise removal adopts the improved mean filtering method, the formula is: where x′ i is the filtered data, n=2k+1 is the filter window size, x j is the original data; i represents the serial number of the data point, k is an integer, and j is the serial number of the data point in the filter window; data interpolation uses linear interpolation, and the formula is where x i is the interpolated data, t i is the corresponding time; x i-1 and x i+1 are respectively i The values ​​of the adjacent previous and next data points, t i-1 and t i+1 Corresponding to x i-1 and x i+1 Time, data normalization maps the data to a specific interval [a, b], the formula is x norm is the normalized data, x min and x max is the minimum and maximum value of the original data; x is a value in the original data;

[0014] Health status analysis module: The pre-processed physiological data and environmental data are analyzed with the help of AI analysis algorithms to determine the user's health status. The health status analysis model is constructed using convolutional neural networks, recurrent neural networks and their variants, long short-term memory networks, and gated recurrent unit networks. The model input is the feature vector of physiological data and environmental data in time series, and the output is the health status category. The model training process is as follows:

[0015] Collect a large number of user data samples labeled with health status categories, and divide the samples into training set, validation set and test set according to the ratio r1:r2:r4;

[0016] Use the training set to train the model and calculate the cross entropy loss function. The formula is: where y c is the true category label, c is the category number, and C is the total number of categories; p c To predict the class probability, the model parameters are adjusted through the back-propagation algorithm;

[0017] Model performance was evaluated on the validation set, and specific disease risk predictions were included, including cardiovascular disease risk prediction, based on heart rate variability in physiological data, calculated as the root mean square of the difference between adjacent heart rate intervals. where RR i is the heart rate interval, N is the total number of heart rate intervals, and the blood pressure fluctuation is calculated as the blood pressure standard deviation BP i is the blood pressure value, n is the number of blood pressure measurements, The average blood pressure value is characterized by combining the altitude and ambient temperature factors in the environmental data and using the logistic regression model for prediction. The specific formula is: Where P is the probability of disease, w0 is the intercept, and w i is the feature weight, i represents the feature number, x i When the predicted risk exceeds the set threshold, an early warning is issued.

[0018] Health information management module: stores collected physiological data, environmental data, analyzed health status and disease risk prediction results, builds user health files, and uses structured data format to store files. Provides users with personalized health advice based on their health status and disease risk prediction results, including diet advice, exercise advice, sleep advice, and medical advice, and connects user health information with third-party medical platforms to achieve data sharing;

[0019] Interaction module: Build an interactive interface between users and smart watches, including a display screen, a touch operation interface, and a voice interaction device. Users can view health information, set monitoring parameters, and receive early warning notifications. When the system issues an early warning, it reminds the user through vibration, sound prompts, and display of eye-catching prompt information. The reminder method can be customized by the user.

[0020] Furthermore, it also includes a motion data precision collection unit: the motion data collection sensor adopts a combination of accelerometer, gyroscope and magnetometer, and the accelerometer measurement accuracy is ±0.1m / s 2 , its measurement range is -2g to 2g, g is the acceleration of gravity; the gyroscope measurement accuracy is ±0.5° / s, and the measurement range is -360° / s to 360° / s, which is used to accurately obtain the rotation information in motion; the magnetometer measurement accuracy is ±10μT, and the measurement range is -100μT to 100μT, which assists in judging the direction of motion. The Kalman filter algorithm is used as the sensor fusion algorithm to fuse the collected raw data and calculate the motion posture and trajectory. The attitude solution algorithm uses the quaternion method to fuse the data of the accelerometer, gyroscope and magnetometer to obtain accurate roll angle, pitch angle and yaw angle, and then determine the motion posture; the motion trajectory calculation is based on the inertial navigation principle, combining the data of the accelerometer and gyroscope, and obtaining the displacement information through integral operation, and then using the magnetometer data for direction correction to finally determine the motion trajectory.

[0021] Furthermore, it also includes a model optimization unit: the deep learning algorithm in the health status analysis module uses transfer learning technology and fine-tunes using a pre-trained model trained on a large-scale medical and health data set; the selection of the pre-trained model is based on its performance in similar health monitoring tasks, including a model architecture trained on a public health data set; during the fine-tuning process, the fully connected layer of the model is retrained based on the characteristics of the collected user data, the learning rate is set to between 0.001 and 0.01, the stochastic gradient descent optimization algorithm is used, the momentum parameter is set to 0.9, and through multiple iterative training, the model is adapted to the distribution characteristics of the system user data.

[0022] Furthermore, the user health records in the health information management module are stored in a local database and a cloud database. The local database uses embedded flash storage technology; the cloud database is based on a distributed storage architecture, including the use of Hadoop distributed file system HDFS or Amazon S3 storage service; data encryption transmission uses the SSL / TLS protocol to encrypt data during data transmission; data storage encryption uses the AES-256 encryption algorithm to encrypt user health information stored in the local database and the cloud database. The cloud database has a data redundancy backup mechanism and adopts a multi-copy storage strategy with 3 to 5 copies distributed in data centers in different geographical locations.

[0023] Furthermore, the voice interaction device in the interaction module supports multiple language recognition, and the voice recognition model adopts end-to-end voice recognition technology based on deep learning, including a convolutional neural network-recurrent neural network CNN-RNN architecture; the voice synthesis technology adopts a voice synthesis model based on deep learning, including a WaveNet or Tacotron model, and the training data set contains voice intonation, rhythm and emotional expression samples.

[0024] Furthermore, the system also includes a firmware upgrade module, which is used to regularly check and update the system firmware version on the server. The checking period can be customized by the user and is once a week by default. When a new version is available, the update package is downloaded via wireless communication. During the firmware upgrade process, the downloaded update package is first checked for integrity, and the hash value of the update package is calculated using the MD5 or SHA-256 hash algorithm and compared with the hash value provided by the server. Then, the differential upgrade technology is used to download and update only the part of the code and data that is different from the current version. During the upgrade process, the system will automatically back up the current firmware version and user data, and the backup data will be stored in the local secure storage area. The backup data will be automatically deleted after the upgrade is successful. If the upgrade fails, the system will automatically restore to the previous firmware version and prompt the user why the upgrade failed.

[0025] Furthermore, the data cleaning in the data preprocessing module also includes marking abnormal data and recording the time and type of abnormal data; the abnormal types of environmental data include too high or too low temperature, abnormal humidity and numerical value. For abnormal data, the system automatically triggers the abnormal analysis program to analyze the cause of the abnormality. If the same type of abnormal data appears multiple times in succession, the system first determines whether it is a sensor failure. Through the built-in sensor self-detection program, the working state of the sensor is detected, including the electrical parameters and signal stability of the sensor; if the sensor failure is ruled out, the user's movement state and emotional state factors are combined to comprehensively analyze whether the abnormal data is caused by the user's abnormal physiological state. If it is determined that the user is in an abnormal physiological state, the system will record the abnormal event in the health information, but will not trigger an early warning; if the cause of the abnormality cannot be determined or the abnormal situation is serious, the system will immediately issue an early warning and upload the abnormal data and related analysis results to the cloud server.

[0026] Furthermore, the health status category in the health status analysis module also covers psychological status assessment, by analyzing heart rate variability, using a time domain analysis method, including calculating the root mean square of the difference between adjacent heart rate intervals where RR i and RR i+1 is the heart rate interval, and N is the total number of heart rate intervals; at the same time, the frequency domain analysis method is combined to calculate the low-frequency to high-frequency power ratio LF / HF, which is used to evaluate the balance state of the autonomic nervous system, sleep data, and exercise data. Combined with the user's daily activity records, machine learning algorithms are used, including support vector machines, and the kernel function uses radial basis function Where γ is the kernel parameter, a i and a j represents two data points in the data set, ||a i -a j || represents the Euclidean distance between the two data points, and the optimal value is selected through cross-validation; decision tree, the C4.5 algorithm is used to build a decision tree model, and the optimal features are selected for classification according to the information gain ratio to evaluate the user's psychological state. Based on the evaluation results, psychological adjustment suggestions are provided to users, including relaxation training methods, progressive muscle relaxation training, and time management suggestions.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] In terms of data collection, the system integrates a variety of high-precision sensors to comprehensively and accurately collect physiological and environmental data, providing a reliable foundation for subsequent analysis.

[0029] The data preprocessing module effectively removes noise and outliers through advanced methods to ensure data quality. Improved mean filtering and other methods ensure data accuracy, while linear interpolation and normalization further optimize the data to make the analysis results more reliable.

[0030] The health status analysis module uses advanced AI algorithms, such as multiple neural networks, to accurately determine health status. Through comprehensive data and scientific algorithms, it can more accurately predict risks such as cardiovascular disease and detect potential health problems at an early stage.

[0031] The health information management module implements local and cloud dual storage, ensuring data security and easy sharing. It connects with third-party medical platforms to facilitate doctors to obtain information for remote diagnosis, while personalized health advice can help users improve their lifestyles.

[0032] The interactive module is designed to be user-friendly, and the diversified interactive modes make it easy for users to operate. The voice interaction function is powerful, supports multiple languages ​​and has high recognition accuracy.

[0033] In addition, the system also has a firmware upgrade module to ensure that the system is always up to date. Units such as precise motion data collection and model optimization further improve system performance. It also has outstanding performance in aspects such as psychological state assessment, protecting user health in all aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a schematic block diagram of a smart watch health information monitoring and management system based on AI technology proposed by the present invention. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0037] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined. In addition, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, and it can be the internal connection of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0038] Reference Figure 1 :A smart watch health information monitoring and management system based on AI technology, including:

[0039] Data acquisition module: Integrates multiple sensors to collect users' various physiological and environmental data in real time. Physiological data covers heart rate, blood pressure, blood oxygen saturation, body temperature, sleep, and exercise data, including number of steps, exercise distance, exercise type, and exercise intensity. Environmental data includes ambient temperature, ambient humidity, altitude, and ultraviolet intensity.

[0040] Data preprocessing module: preprocessing the collected physiological data and environmental data. In the data cleaning phase, outliers and noise are removed. The noise removal adopts the improved mean filtering method. The formula is: where x′ i is the filtered data, n=2k+1 is the filter window size, x j is the original data; i represents the serial number of the data point, k is an integer, which determines the number of adjacent data points involved in the filtering calculation, and j is the serial number of the data point in the filtering window; data interpolation adopts linear interpolation method, and the formula is where x i is the interpolated data, t i is the corresponding time; x i-1 and x i+1 are respectively i The values ​​of the adjacent previous and next data points, t i-1 and t i+1 Corresponding to x i-1 and x i+1 Time, data normalization maps the data to a specific interval [a, b], the formula is x norm is the normalized data, x min and xmax are the minimum and maximum values ​​of the original data; x is a value in the original data.

[0041] Health status analysis module: Use AI analysis algorithms to conduct in-depth analysis of pre-processed physiological data and environmental data to determine the user's health status. Use deep learning algorithms (such as convolutional neural networks, recurrent neural networks and their variants, long short-term memory networks, and gated recurrent unit networks) to build a health status analysis model. The model input is the feature vector of physiological data and environmental data in a time series, and the output is a health status category (such as healthy, sub-healthy, and disease risk). The model training process is as follows: Collect a large number of user data samples labeled with health status categories, and divide the samples into training set, validation set, and test set according to the ratio of r1:r2:r4. Use the training set to train the model and calculate the cross entropy loss function. The formula is: where y c is the true category label, c is the category number, and C is the total number of categories; p c To predict the category probability, the model parameters are adjusted through the back propagation algorithm, and the model performance is evaluated on the validation set (the indicators include accuracy, recall rate, and F1 value. When the performance no longer improves, the training is stopped, and the generalization ability of the model is finally tested on the test set. It also includes specific disease risk prediction, such as cardiovascular disease risk prediction, based on the heart rate variability in physiological data, and the calculation method is the root mean square difference between adjacent heart rate intervals. where RR i is the heart rate interval, N is the total number of heart rate intervals, and the blood pressure fluctuation is calculated as the standard deviation of blood pressure over a period of time. BP i is the blood pressure value, n is the number of blood pressure measurements, is the average blood pressure value) and other characteristics, combined with factors such as altitude (high altitude areas increase the risk of cardiovascular disease) and ambient temperature (extreme temperature affects the cardiovascular system) in environmental data, a logistic regression model is used for prediction. The specific formula is: Where P is the probability of disease, w0 is the intercept, and w i is the feature weight, i represents the feature number, x i When the predicted risk exceeds the set threshold (such as the cardiovascular disease risk threshold is 0.7), an early warning is issued.

[0042] Health information management module: Stores collected physiological data, environmental data, analyzed health status and disease risk prediction results, builds user health records, and uses structured data format (such as JSON format) for easy data query and retrieval. Provides personalized health advice to users based on their health status and disease risk prediction results, such as dietary advice (recommended diet plans based on health status, such as low-salt diets for hypertensive patients and low-sugar diets for diabetic patients), exercise advice (recommended exercise methods, intensity and time based on physical condition and exercise data, such as at least 150 minutes of moderate-intensity aerobic exercise per week for sub-healthy people), sleep advice (based on sleep data analysis results, such as insufficient sleep duration or poor quality, it is recommended to adjust work and rest time, improve sleep environment, etc.), and medical advice (recommends timely medical treatment and provides information on nearby medical institutions when the risk of disease is high). Connects user health information with third-party medical platforms (such as hospital information systems and telemedicine platforms) to achieve data sharing, help doctors obtain user health information, and provide support for remote diagnosis and treatment.

[0043] Interaction module: Build an interactive interface between users and smart watches, including display screen, touch operation interface, and voice interaction device. Users can view health information (such as real-time physiological data, health status analysis results, health advice, etc.), set monitoring parameters (such as data collection frequency, warning threshold, etc.), and receive warning notifications. When the system issues a warning, it reminds the user through vibration, sound prompts, display of eye-catching prompt information, etc. The reminder method can be customized by the user.

[0044] The present invention also includes a motion data precision acquisition unit: the motion data acquisition sensor adopts a combination of an accelerometer, a gyroscope and a magnetometer, and the accelerometer measurement accuracy is ±0.1m / s 2 , its measurement range is -2g to 2g (g is the acceleration of gravity), which can accurately sense the acceleration changes under different motion states; the gyroscope has a measurement accuracy of ±0.5° / s, and the measurable range is -360° / s to 360° / s, which is used to accurately obtain the rotation information in motion; the magnetometer has a measurement accuracy of ±10μT, and the measurement range is -100μT to 100μT, which can assist in determining the direction of motion. The Kalman filter algorithm is used as a sensor fusion algorithm to fuse the collected raw data and calculate the motion posture and trajectory. The attitude solution algorithm uses the quaternion method to fuse the data of the accelerometer, gyroscope and magnetometer to obtain accurate roll angle, pitch angle and yaw angle, and then determine the motion posture; the motion trajectory calculation is based on the inertial navigation principle, combined with the data of the accelerometer and gyroscope, and the displacement information is obtained through integral operation, and the magnetometer data is used for direction correction to finally determine the motion trajectory.

[0045] The present invention also includes a model optimization unit: the deep learning algorithm in the health status analysis module uses transfer learning technology and fine-tunes using a pre-trained model trained on a large-scale medical and health data set. The selection of the pre-trained model is based on its good performance in similar health monitoring tasks, such as a model architecture with high accuracy obtained by training on a large public health data set. During the fine-tuning process, the fully connected layer of the model is retrained according to the characteristics of the user data collected by this system, the learning rate is set to between 0.001 and 0.01, and the stochastic gradient descent optimization algorithm is used. The momentum parameter is set to 0.9. Through multiple iterative training, the model gradually adapts to the distribution characteristics of the user data of this system, thereby reducing training time and data requirements and improving model accuracy.

[0046] In the present invention, the user health records in the health information management module are stored in a local database and a cloud database. The local database adopts embedded flash storage technology, with a storage capacity of 1GB to 4GB, fast read and write capabilities, and a data reading speed of 50MB / s to 100MB / s, which is used for fast data access to meet the local real-time data query requirements of smart watches; the cloud database is based on a distributed storage architecture, such as Hadoop distributed file system (HDFS) or Amazon S3 storage service, and the storage capacity can be elastically expanded according to the number of users and data growth. The initial configuration is 100GB to 1TB, which is used for data backup and multi-device synchronization. Data encryption transmission adopts SSL / TLS protocol to encrypt data during data transmission to ensure data transmission security; data storage encryption adopts AES-256 encryption algorithm to encrypt user health information stored in the local database and cloud database to ensure data storage security. At the same time, the cloud database has a data redundancy backup mechanism, adopts a multi-copy storage strategy, and the number of copies is 3 to 5, distributed in data centers in different geographical locations to prevent data loss and ensure the integrity and availability of user health information.

[0047] In the present invention, the voice interaction device in the interactive module supports multiple language recognition (no less than 5 languages, such as Chinese, English, Japanese, Korean, French, etc.), and the voice recognition model adopts end-to-end voice recognition technology based on deep learning, such as convolutional neural network-recurrent neural network (CNN-RNN) architecture, and the training data set contains voice samples from different languages, different accents and speaking speeds, with a number of no less than 100,000. The voice recognition accuracy is not less than 90%, and the accuracy can reach more than 95% in a quiet environment, and the accuracy is not less than 90% in a noisy environment (signal-to-noise ratio is not less than 10dB). Speech synthesis technology uses a deep learning-based speech synthesis model, such as the WaveNet or Tacotron model. The training data set contains rich samples of speech intonation, rhythm, and emotional expression. The naturalness of speech synthesis meets certain standards (such as an average opinion score of not less than 3.5), and can generate natural, fluent, and emotional speech output. It supports personalized adjustment of speech speed, intonation, and volume. The adjustment range is 0.5 to 2 times the normal speed, -50% to 50% of the normal intonation, and 0.5 to 1.5 times the normal volume.

[0048] In the present invention, the system also includes a firmware upgrade module for regularly checking and updating the system firmware version on the server. The inspection cycle can be customized by the user and is set by default once a week. When there is a new version, the update package is downloaded through wireless communication (such as Bluetooth 5.0 and above, the transmission rate can reach 2Mbps to 4Mbps; Wi-Fi uses 802.11n or higher standards, the transmission rate can reach 150Mbps to 600Mbps). During the firmware upgrade process, the downloaded update package is firstly checked for integrity, and the hash value of the update package is calculated using the MD5 or SHA-256 hash algorithm, and compared with the hash value provided by the server to ensure that the update package data is complete and correct. Then, the differential upgrade technology is used to download and update only the part of the code and data that is different from the current version, reducing the upgrade data volume and upgrade time. During the upgrade process, the system will automatically back up the current firmware version and user data, and the backup data will be stored in the local secure storage area. The backup data will be automatically deleted after the upgrade is successful; if the upgrade fails, the system will automatically restore to the previous firmware version, and prompt the user why the upgrade failed to ensure data security and system stability. It supports both online and offline upgrades. The offline upgrade package can be downloaded from a computer and then transferred to the smart watch for upgrade operations.

[0049] In the present invention, the data cleaning in the data preprocessing module also includes marking abnormal data, recording the time (accurate to seconds), type (such as abnormal physiological data types including too high or too low heart rate, abnormal blood pressure fluctuations, etc.; abnormal environmental data types including too high or too low temperature, abnormal humidity, etc.) and value of abnormal data. For abnormal data, the system automatically triggers the abnormal analysis program to analyze the cause of the abnormality. If the same type of abnormal data appears multiple times in succession (such as abnormal heart rate for more than 5 times in a row), the system first determines whether it is a sensor failure, and detects the working state of the sensor through the built-in sensor self-detection program, including the electrical parameters of the sensor, signal stability, etc.; if the sensor failure is ruled out, the user's motion state (judged by motion sensor data), emotional state (if the smart watch integrates the heart rate variability analysis function, the emotional state can be inferred based on this) and other factors are combined to comprehensively analyze whether the abnormal data is caused by the user's abnormal physiological state (such as strenuous exercise, emotional excitement, etc.). If it is determined that the user is in an abnormal physiological state, the system will record the abnormal event in the health information, but will not trigger an early warning; if the cause of the abnormality cannot be determined or the abnormality is more serious (such as severe arrhythmia, etc.), the system will immediately issue an early warning and upload the abnormal data and related analysis results to the cloud server for further analysis and processing by professionals.

[0050] In the present invention, the health status category in the health status analysis module also covers psychological status assessment. By analyzing heart rate variability (using time domain analysis methods, such as calculating the root mean square difference between adjacent heart rate intervals), where RR i and RR i+1 is the heart rate interval, N is the total number of heart rate intervals; combined with the frequency domain analysis method, the low-frequency to high-frequency power ratio LF / HF is calculated to evaluate the balance state of the autonomic nervous system), sleep data (such as sleep stage distribution, by analyzing brain wave signals to determine the duration of wakefulness, light sleep, deep sleep and rapid eye movement sleep stages; sleep cycle, calculate the number and duration of complete sleep cycles), exercise data (such as exercise frequency, counting the number of exercise days per week and the number of exercises per day; exercise regularity, using time series analysis methods to determine the regularity of exercise time and intensity) and other data, combined with users' daily activity records (such as the time and type of mobile phone application usage, obtained through connection with smartphones; social activity frequency, based on call records, text messages and social software usage analysis), using machine learning algorithms (such as support vector machines, the kernel function uses radial basis function Where γ is the kernel parameter, a i and a j represents two data points in the data set, ||a i -a j|| represents the Euclidean distance between the two data points, and the optimal value is selected through cross-validation; decision tree, the decision tree model is constructed using the C4.5 algorithm, and the optimal feature is selected for classification based on the information gain ratio) to evaluate the user's psychological state (such as stress level, which is divided into three levels: low, medium, and high; emotional state, including positive, negative, and neutral, etc.). Based on the evaluation results, psychological adjustment suggestions are provided to users, such as relaxation training methods (such as deep breathing exercises, guiding users to inhale and exhale according to a specific rhythm, each exercise lasting 5 to 10 minutes; progressive muscle relaxation training, sequentially tensing and relaxing muscles in various parts of the body), time management suggestions (analyzing the rationality of time allocation based on the user's daily activity records, and providing suggestions for reasonable arrangement of work, study, and rest time), etc.

[0051] The above are only preferred specific implementation modes of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A smart watch health information monitoring and management system based on AI technology, characterized in that: include: Data acquisition module: integrates multiple sensors to collect multiple physiological data and environmental data of users in real time; Data preprocessing module: preprocessing the collected physiological data and environmental data, data cleaning, removing outliers and noise. The noise removal adopts the improved mean filtering method, the formula is: where x′ i is the filtered data, n=2k+1 is the filter window size, x j is the original data; i represents the serial number of the data point, k is an integer, and j is the serial number of the data point in the filter window; data interpolation uses linear interpolation, and the formula is where x i is the interpolated data, t i is the corresponding time; x i-1 and x i+1 are respectively i The values ​​of the adjacent previous and next data points, t i-1 and t i+1 Corresponding to x i-1 and x i+1 Time, data normalization maps the data to a specific interval [a, b], the formula is x norm is the normalized data, x min and x max is the minimum and maximum value of the original data; x is a value in the original data; Health status analysis module: The pre-processed physiological data and environmental data are analyzed with the help of AI analysis algorithms to determine the user's health status. The health status analysis model is constructed using convolutional neural networks, recurrent neural networks and their variants, long short-term memory networks, and gated recurrent unit networks. The model input is the feature vector of physiological data and environmental data in time series, and the output is the health status category. The model training process is as follows: Collect a large number of user data samples labeled with health status categories, and divide the samples into training set, validation set and test set according to the ratio r1:r2:r4; Use the training set to train the model and calculate the cross entropy loss function. The formula is: where y c is the true category label, c is the category number, and C is the total number of categories; p c To predict the class probability, the model parameters are adjusted through the back-propagation algorithm; Model performance was evaluated on the validation set, and specific disease risk predictions were included, including cardiovascular disease risk prediction, based on heart rate variability in physiological data, calculated as the root mean square of the difference between adjacent heart rate intervals. where RR i is the heart rate interval, N is the total number of heart rate intervals, and the blood pressure fluctuation is calculated as the blood pressure standard deviation BP i is the blood pressure value, n is the number of blood pressure measurements, The average blood pressure value is characterized by combining the altitude and ambient temperature factors in the environmental data and using the logistic regression model for prediction. The specific formula is: Where P is the probability of disease, w0 is the intercept, and w i is the feature weight, i represents the feature number, x i When the predicted risk exceeds the set threshold, an early warning is issued. Health information management module: stores collected physiological data, environmental data, analyzed health status and disease risk prediction results, builds user health files, and uses structured data format to store files. Provides users with personalized health advice based on their health status and disease risk prediction results, including diet advice, exercise advice, sleep advice, and medical advice, and connects user health information with third-party medical platforms to achieve data sharing; Interaction module: Build an interactive interface between users and smart watches, including a display screen, a touch operation interface, and a voice interaction device. Users can view health information, set monitoring parameters, and receive early warning notifications. When the system issues an early warning, it reminds the user through vibration, sound prompts, and display of eye-catching prompt information. The reminder method can be customized by the user.

2. The AI-based smart watch health information monitoring and management system according to claim 1, characterized in that: It also includes a motion data precision collection unit: the motion data collection sensor uses a combination of accelerometer, gyroscope and magnetometer, and the accelerometer measurement accuracy is ±0.1m / s 2 , its measurement range is -2g to 2g, g is the acceleration of gravity; the gyroscope has a measurement accuracy of ±0.5° / s, and a measurement range of -360° / s to 360° / s, which is used to accurately obtain rotation information in motion; the magnetometer has a measurement accuracy of ±10μT, and a measurement range of -100μT to 100μT, which assists in determining the direction of motion. The Kalman filter algorithm is used as a sensor fusion algorithm to fuse the collected raw data and calculate the motion posture and trajectory. The attitude solution algorithm uses the quaternion method to fuse the data of the accelerometer, gyroscope and magnetometer to obtain accurate roll angle, pitch angle and yaw angle, and then determine the motion posture; The motion trajectory calculation is based on the principle of inertial navigation. It combines the data of the accelerometer and gyroscope, obtains the displacement information through integration operation, and then uses the magnetometer data to correct the direction to finally determine the motion trajectory.

3. The AI-based smart watch health information monitoring and management system according to claim 1, characterized in that: It also includes a model optimization unit: the deep learning algorithm in the health status analysis module uses transfer learning technology to fine-tune using a pre-trained model trained on a large-scale medical and health data set; the pre-trained model is selected based on its performance in similar health monitoring tasks, including a model architecture trained on a public health data set; During the fine-tuning process, the fully connected layer of the model was retrained based on the characteristics of the collected user data. The learning rate was set between 0.001 and 0.01, the stochastic gradient descent optimization algorithm was used, and the momentum parameter was set to 0.

9. Through multiple iterative training, the model was adapted to the distribution characteristics of the system user data.

4. The AI-based smart watch health information monitoring and management system according to claim 1, characterized in that: The user health records in the health information management module are stored in a local database and a cloud database. The local database uses embedded flash storage technology; the cloud database is based on a distributed storage architecture, including the use of Hadoop distributed file system HDFS or Amazon S3 storage service; data encryption transmission uses the SSL / TLS protocol to encrypt data during data transmission; data storage encryption uses the AES-256 encryption algorithm to encrypt user health information stored in the local database and the cloud database. The cloud database has a data redundancy backup mechanism and adopts a multi-copy storage strategy with 3 to 5 copies distributed in data centers in different geographical locations.

5. The AI-based smart watch health information monitoring and management system according to claim 1, characterized in that: The voice interaction device in the interaction module supports multiple language recognition, and the voice recognition model adopts end-to-end voice recognition technology based on deep learning, including a convolutional neural network-recurrent neural network CNN-RNN architecture; the voice synthesis technology adopts a voice synthesis model based on deep learning, including a WaveNet or Tacotron model, and the training data set contains voice intonation, rhythm and emotional expression samples.

6. The AI-based smart watch health information monitoring and management system according to claim 1, characterized in that: The system also includes a firmware upgrade module, which is used to regularly check and update the system firmware version on the server. The check period can be customized by the user and is set once a week by default. When a new version is available, the update package is downloaded via wireless communication. During the firmware upgrade process, the integrity of the downloaded update package is first checked, and the hash value of the update package is calculated using the MD5 or SHA-256 hash algorithm and compared with the hash value provided by the server. Then, the differential upgrade technology is used to download and update only the part of the code and data that is different from the current version. During the upgrade process, the system will automatically back up the current firmware version and user data. The backup data will be stored in the local secure storage area and automatically deleted after the upgrade is successful. If the upgrade fails, the system automatically reverts to the previous firmware version and prompts the user why the upgrade failed.

7. The AI-based smart watch health information monitoring and management system according to claim 1, characterized in that: The data cleaning in the data preprocessing module also includes marking abnormal data and recording the time and type of abnormal data. Abnormal types of environmental data include too high or too low temperature, abnormal humidity and numerical value. For abnormal data, the system automatically triggers the abnormal analysis program to analyze the cause of the abnormality. If the same type of abnormal data appears multiple times in succession, the system first determines whether it is a sensor failure. Through the built-in sensor self-detection program, the working status of the sensor is detected, including the electrical parameters and signal stability of the sensor. If the sensor failure is ruled out, the user's movement status and emotional state factors are combined to comprehensively analyze whether the abnormal data is caused by the user's abnormal physiological state. If it is determined that the user is in an abnormal physiological state, the system will record the abnormal event in the health information, but will not trigger an early warning. If the cause of the abnormality cannot be determined or the abnormal situation is serious, the system will immediately issue an early warning and upload the abnormal data and related analysis results to the cloud server.

8. The AI-based smart watch health information monitoring and management system according to claim 1, characterized in that: The health status category in the health status analysis module also covers psychological status assessment, by analyzing heart rate variability, using time domain analysis methods, including calculating the root mean square of the difference between adjacent heart rate intervals where RR i and RR i+1 is the heart rate interval, and N is the total number of heart rate intervals; at the same time, the frequency domain analysis method is combined to calculate the low-frequency to high-frequency power ratio LF / HF, which is used to evaluate the balance state of the autonomic nervous system, sleep data, and exercise data. Combined with the user's daily activity records, machine learning algorithms are used, including support vector machines, and the kernel function uses radial basis function Where γ is the kernel parameter, a i and a j represents two data points in the data set, ||a i -a j || represents the Euclidean distance between the two data points, and the optimal value is selected through cross-validation; decision tree, the C4.5 algorithm is used to build a decision tree model, and the optimal features are selected for classification according to the information gain ratio to evaluate the user's psychological state. Based on the evaluation results, psychological adjustment suggestions are provided to users, including relaxation training methods, progressive muscle relaxation training, and time management suggestions.

Images