Diet and exercise recommendation method and system based on digital twinning

Abstract

The application provides a diet and exercise recommendation method and system based on digital twinning. The diet and exercise recommendation method based on digital twinning comprises the following steps: constructing a digital twinning model by using collected initial physiological data and dynamic physiological data of a user; simulating and predicting initial physiological data and dynamic physiological data of a registered target user by using the digital twinning model, obtaining a diet and running type matched with the initial physiological data and dynamic physiological data of the registered target user; and recommending a diet and exercise scheme to the registered target user from a database according to the diet and running type matched with the initial physiological data and dynamic physiological data of the registered target user. The system comprises modules corresponding to the method steps.

Description

Technical Field

[0001] This invention proposes a method and system for recommending diet and exercise based on digital twins, belonging to the field of exercise and diet data recommendation technology. Background Technology

[0002] With the accelerating pace of modern society and a significant increase in public awareness of health, personalized health management programs are increasingly becoming a key element in meeting people's health needs. Traditionally, dietary and exercise recommendations have often relied on universally applicable health manuals or general guidelines based on experience. These methods fail to fully consider the immediate impact of each individual's unique physiological structure, lifestyle habits, and daily environment on their health status.

[0003] Specifically, each person's physical condition, metabolic rate, genetic background, and daily activity level differ, and these factors are crucial for developing an effective health management plan. Meanwhile, changes in the external environment, such as seasonal changes, work stress, sleep quality, and even psychological state, also have a significant impact on an individual's health. However, traditional recommendation methods often overlook these subtle individual differences and dynamic changes, significantly reducing the relevance and effectiveness of the advice. Therefore, developing a diet and exercise recommendation system that can accurately match individual needs and dynamically adjust recommendations is particularly important. Summary of the Invention

[0004] This invention provides a method and system for recommending diet and exercise based on digital twins, in order to solve the problems existing in the prior art. The technical solution adopted is as follows:

[0005] A digital twin-based diet and exercise recommendation method, comprising:

[0006] Digital twin models are constructed using the initial and dynamic physiological data of users that have been collected;

[0007] After the digital twin model is built, initial physiological data is collected from the registered target users, and the dynamic physiological data of the target users is dynamically adjusted and collected according to the dynamic changes of the target users' dynamic physiological data to obtain the initial physiological data and dynamic physiological data of the registered target users.

[0008] The initial and dynamic physiological data of the registered target users are simulated and predicted using a digital twin model to obtain diet and exercise plans that match the diet and exercise types of the registered target users, and then recommended to the registered target users.

[0009] Furthermore, a digital twin model is constructed using the collected initial and dynamic physiological data of users, including:

[0010] The system retrieves initial and dynamic physiological data of users from the database; the initial physiological data includes age, gender, weight, height, and basal metabolic rate; the dynamic physiological data includes heart rate, blood pressure, and calories burned during exercise.

[0011] The collected initial and dynamic physiological data of users are cleaned to remove invalid, duplicate, and erroneous data, and preprocessed initial and dynamic physiological data are obtained.

[0012] A digital twin model is constructed using the preprocessed initial physiological data and dynamic physiological data. The constructed digital twin model is then trained and validated to obtain the completed digital twin model.

[0013] Furthermore, the structure of the digital twin model is as follows:

[0014] The input layer is used to input the initial physiological data and dynamic physiological data of the target user, and to perform unified and standardized processing on the initial physiological data and dynamic physiological data of the target user.

[0015] The multi-scale time window layer includes multiple parallel LSTM sub-networks, and each LSTM sub-network corresponds to a time window; wherein, the parallel LSTM sub-networks are used to process input sequences of different lengths to control the digital twin model to capture data patterns at different time scales.

[0016] An adaptive gated fusion layer is used to dynamically adjust the parameter values ​​corresponding to the importance and contribution of each LSTM sub-network based on the current input initial physiological data and dynamic physiological data, and to synthesize the output of the dynamically adjusted LSTM sub-networks to form a unified hidden layer.

[0017] The context-aware fusion layer combines the output of the adaptive gated fusion layer with contextual information from the target user; this allows for dynamic adjustment of the model's response, making the output more personalized and accurate.

[0018] The output layer is used to output the diet and run type.

[0019] Furthermore, the steps for obtaining the time window are as follows:

[0020] Retrieve the preset base decay rate from the database;

[0021] Real-time monitoring of the frequency of changes in collected dynamic physiological data;

[0022] The dynamic attenuation factor corresponding to each LSTM subnetwork is obtained by using the preset basic attenuation rate and the data change frequency of dynamic physiological data.

[0023] The time window length corresponding to each LSTM sub-network is obtained using the dynamic decay factor; wherein, the time window length corresponding to each LSTM sub-network is obtained by the following formula:

[0024]

[0025] Among them, K i α(i, β, λ) represents the time window length corresponding to the i-th LSTM subnetwork; T represents the preset observation period length; α(i, β, λ) represents the dynamic decay factor corresponding to the i-th LSTM subnetwork; N represents the total number of LSTM subnetworks; w i α represents the preset initial weight value corresponding to the i-th LSTM subnetwork; α0 represents the preset base decay rate; β represents the frequency factor of dynamic physiological data change; i represents the index corresponding to each LSTM subnetwork; λ represents the preset periodic intensity value of dynamic physiological data.

[0026] Furthermore, the dynamic attenuation factor corresponding to each LSTM sub-network is obtained using the preset base attenuation rate and the data change frequency of dynamic physiological data, including:

[0027] The frequency of data changes in the collected dynamic physiological data;

[0028] The frequency of change factor is obtained using the frequency of change in the dynamic physiological data; wherein, the frequency of change factor is obtained by the following formula:

[0029]

[0030] Where β represents the frequency factor of dynamic physiological data; r represents the adjustment parameter, and the value range of the adjustment parameter is [0.1, 0.9]; f 01 f 02 and f 03 These correspond to the percentage changes in heart rate, blood pressure, and calories burned during exercise, respectively, representing the maximum range of change. p01 v p02 and v p03 These correspond to the average rate of change of heart rate, blood pressure, and exercise calories consumed per unit time, respectively.

[0031] The dynamic attenuation factor corresponding to each LSTM sub-network is obtained by combining the changing frequency factor with the basic attenuation rate.

[0032] The dynamic attenuation factor for each LSTM sub-network is obtained using the following formula:

[0033]

[0034] Where α(i, β, λ) represents the dynamic decay factor corresponding to the i-th LSTM subnetwork; α0 represents the preset base decay rate; β represents the frequency factor of dynamic physiological data change; i represents the index corresponding to each LSTM subnetwork, and i = 1, 2, 3, ..., N; N represents the total number of LSTM subnetworks; and λ represents the preset periodic intensity value of dynamic physiological data.

[0035] Furthermore, after the digital twin model is constructed, initial physiological data is collected from registered target users. The dynamic physiological data of the target users is then dynamically adjusted based on their dynamic changes, resulting in the acquisition of both initial and dynamic physiological data for the registered target users, including:

[0036] After the digital twin model is constructed, an initial physiological data input instruction is sent to the registered target user, and the initial physiological data input by the registered target user is received in real time; wherein, the initial physiological data includes age, gender, weight, height and basal metabolic rate;

[0037] The system collects real-time dynamic physiological data of registered target users through wearable devices; wherein the dynamic physiological data includes heart rate, blood pressure, and calories burned during exercise.

[0038] The frequency of dynamic physiological data collection is dynamically adjusted by extracting the frequency of change of the dynamic physiological data of the registered target users.

[0039] Furthermore, the frequency of dynamic physiological data collection is dynamically adjusted based on the frequency of change of the registered target users' dynamic physiological data, including:

[0040] Extract the frequency of changes in heart rate, blood pressure, and exercise calories burned from the dynamic physiological data of the registered target users;

[0041] The frequency parameters corresponding to the changes in heart rate, blood pressure, and exercise calories are obtained using the frequency of change of these parameters.

[0042] The acquisition frequency of the dynamic physiological data is adjusted using the frequency parameters corresponding to the changes in heart rate, blood pressure, and exercise calories burned. The adjusted acquisition frequency of the dynamic physiological data is obtained using the following formula:

[0043]

[0044] Among them, f new Indicates the adjusted frequency of dynamic physiological data acquisition; f y ξ represents the frequency of dynamic physiological data collection before adjustment; ξ represents the preset adjustment coefficient, with a value range of 0.26-0.41; X 01 X represents the frequency parameter corresponding to the change in heart rate; 02 The parameter representing the frequency of change in blood pressure; X 03 This represents the frequency parameter corresponding to the change in calories burned during exercise.

[0045] Further, the frequency parameters corresponding to the changes in heart rate, blood pressure, and exercise calories burned are obtained using the frequency of these changes, including:

[0046] The frequency of heart rate changes contained in the dynamic physiological data is used to obtain the frequency parameter corresponding to the heart rate; wherein, the frequency parameter corresponding to the heart rate is obtained by the following formula:

[0047]

[0048] Among them, X 01 X represents the frequency parameter corresponding to the change in heart rate; b01 This represents the standard deviation of heart rate; m 01 Indicates the number of heart rate measurements; H i H represents the heart rate value corresponding to the i-th data collection; i+1 H represents the heart rate value corresponding to the (i+1)th data collection; f This indicates the maximum permissible fluctuation range of the heart rate value;

[0049] The blood pressure change frequency parameter is obtained by utilizing the blood pressure change frequency contained in the dynamic physiological data; wherein, the blood pressure change frequency parameter is obtained by the following formula:

[0050]

[0051] Among them, X 02 The parameter representing the frequency of change in blood pressure; X x02 X represents the standard deviation of diastolic blood pressure; y02 This represents the standard deviation of systolic blood pressure; m 02 Indicates the number of blood pressure measurements; S 01i+1 and S 01i S represents the diastolic blood pressure values ​​corresponding to the (i+1)th and ithth measurements; 02i+1 and S 02i This represents the systolic blood pressure values ​​collected in the (i+1)th and ithth measurements.

[0052] The frequency of change of exercise-induced calorie expenditure contained in the dynamic physiological data is used to obtain the frequency parameter corresponding to the exercise-induced calorie expenditure; wherein, the frequency parameter corresponding to the exercise-induced calorie expenditure is obtained by the following formula:

[0053]

[0054] Among them, X 03 X represents the frequency parameter corresponding to the change in calories burned during exercise; b03 The standard deviation of calories burned during exercise; m 03 E represents the number of times calories burned during exercise. i E represents the calorie expenditure value corresponding to the i-th data collection. i+1 E represents the calorie expenditure value corresponding to the (i+1)th data collection. f This indicates the preset reference value for the fluctuation of exercise calories.

[0055] Furthermore, using a digital twin model, the initial and dynamic physiological data of the registered target users are simulated and predicted to obtain diet and exercise plans that match the diet and exercise types of the registered target users, and recommendations are made to the registered target users, including:

[0056] The initial and dynamic physiological data of the registered target users are simulated and predicted using a digital twin model to obtain diet and exercise types that match the initial and dynamic physiological data of the registered target users.

[0057] Based on the diet and exercise types that match the initial and dynamic physiological data of the registered target users, diet and exercise plans are retrieved from the database and recommended to the registered target users.

[0058] A digital twin-based diet and exercise recommendation system, comprising:

[0059] The digital twin model building module is used to build digital twin models using the initial and dynamic physiological data of users that have been collected.

[0060] The data acquisition module is used to collect initial physiological data of registered target users after the digital twin model is built, and to dynamically adjust the collection of dynamic physiological data of target users according to the dynamic changes of dynamic physiological data of target users, so as to obtain the initial physiological data and dynamic physiological data of registered target users.

[0061] The program recommendation module is used to simulate and predict the initial and dynamic physiological data of the registered target users using a digital twin model, obtain diet and exercise programs that match the diet and exercise types of the registered target users, and recommend them to the registered target users.

[0062] Beneficial effects of this invention:

[0063] The proposed method and system for diet and exercise recommendations based on digital twins are based on a digital twin model constructed from the user's initial and dynamic physiological data. This model accurately reflects each user's physiological characteristics and trends, providing a highly personalized foundation for subsequent health management applications. Through data cleaning and model training and validation, the accuracy and reliability of the digital twin model are effectively improved. The model can more accurately simulate and predict the user's physiological state and its dynamic changes, providing more scientific health management advice. The digital twin model can receive and process the user's dynamic physiological data in real time, dynamically adjusting health management recommendations based on changes in the user's physiological state. This dynamic adaptability makes health management services more aligned with the user's actual needs, improving the service's relevance and effectiveness. Health management applications based on digital twin models can provide strong decision support for professionals such as doctors and health managers. Through model simulation and prediction results, professionals can more accurately assess the user's health status, predict potential risks, and develop targeted health management plans. Personalized health management services and dynamically adaptive recommendations significantly enhance the user experience. Users can obtain tailored health management plans based on their actual situation, better manage their health, and improve their quality of life. Attached Figure Description

[0064] Figure 1 This is a flowchart of the method described in this invention;

[0065] Figure 2 This is a system block diagram of the system described in this invention. Detailed Implementation

[0066] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0067] This invention proposes a diet and exercise recommendation method based on digital twins, such as... Figure 1 As shown, the diet and exercise recommendation method based on digital twins includes:

[0068] S1. Construct a digital twin model using the initial and dynamic physiological data of the users that have been collected;

[0069] S2. After the digital twin model is constructed, initial physiological data is collected for the registered target users, and the dynamic physiological data of the target users is dynamically adjusted and collected according to the dynamic changes of the dynamic physiological data of the target users, so as to obtain the initial physiological data and dynamic physiological data of the registered target users.

[0070] S3. Using a digital twin model, simulate and predict the initial and dynamic physiological data of the registered target users to obtain diet and exercise plans that match the diet and exercise types of the registered target users, and recommend them to the registered target users.

[0071] The working principle of the above technical solution is as follows: Historical physiological data of users is retrieved from the database, including static initial physiological data (such as age, gender, weight, height, and basal metabolic rate) and dynamic physiological data (such as heart rate, blood pressure, and exercise calories burned). This data forms the foundation for building the digital twin model. Cleaning the retrieved data is a crucial data quality control process. By removing invalid data (such as data with too many missing values ​​or data that is clearly contrary to common sense), duplicate data, and erroneous data, the accuracy and reliability of subsequent model construction are ensured. The preprocessed data is used to build the digital twin model. The completed digital twin model is then trained and validated. The training process involves adjusting the model parameters using a large amount of data to enable the model to more accurately reflect actual conditions; the validation process evaluates the model's performance on unseen data to ensure the model's generalization ability and stability.

[0072] The above technical solution achieves the following results: A digital twin model, constructed based on the user's initial and dynamic physiological data, accurately reflects each user's physiological characteristics and trends, providing a highly personalized foundation for subsequent diet and exercise recommendations. Through data cleaning and model training validation, the accuracy and reliability of the digital twin model are effectively improved, enabling it to more accurately predict the user's physiological responses under different diet and exercise programs. The model can receive and process the user's dynamic physiological data in real time, dynamically adjusting diet and exercise recommendations based on changes in the user's physiological state, ensuring the scientific validity and timeliness of the suggestions. The diet and exercise recommendation method based on the digital twin model can provide users with more precise and effective health management services, helping them better manage their health, prevent disease, and improve their quality of life.

[0073] In summary, the above technical solutions achieve precise management and effective intervention of users' health status by constructing highly personalized digital twin models and combining them with dynamic physiological data to update recommendation schemes in real time.

[0074] One embodiment of the present invention utilizes collected initial and dynamic physiological data of users to construct a digital twin model, including:

[0075] S101. Retrieve the user's initial physiological data and dynamic physiological data collected from the database; wherein, the initial physiological data includes age, gender, weight, height, and basal metabolic rate; the dynamic physiological data includes heart rate, blood pressure, and exercise-induced calorie expenditure;

[0076] S102. Perform data cleaning on the collected initial physiological data and dynamic physiological data of the user to remove invalid data, duplicate data and erroneous data, and obtain preprocessed initial physiological data and dynamic physiological data;

[0077] S103. Construct a digital twin model using the preprocessed initial physiological data and dynamic physiological data, and train and validate the constructed digital twin model to obtain the completed trained and validated digital twin model.

[0078] The working principle of the above technical solution is as follows: Initial and dynamic physiological data of the user are retrieved from a pre-established database. This data forms the basis for building the digital twin model. Initial physiological data includes static information that does not change frequently, such as age, gender, weight, height, and basal metabolic rate; dynamic physiological data includes information that changes in real time, such as heart rate, blood pressure, and calories burned during exercise. Data cleaning is performed on the collected raw data, a crucial step to ensure the accuracy of the model. Data cleaning includes removing invalid data (such as data with too many missing values ​​or obviously abnormal data points), duplicate data (identical data resulting from system errors or repeated entry), and erroneous data (such as data that clearly does not conform to physiological common sense). Through this technical solution, clean, accurate, and reliable preprocessed data can be obtained, providing a high-quality data source for subsequent model construction. Using the preprocessed data, combined with the principles and methods of digital twin technology, a digital twin model that reflects the user's physiological state and its dynamic changes is constructed. This process involves selecting appropriate mathematical models, algorithms, and parameters to achieve accurate simulation and prediction of the user's physiological characteristics. The completed digital twin model is then trained and validated. The training process involves adjusting model parameters using a large amount of historical data to enable the model to more accurately reflect the user's physiological state and its changing trends. The validation process evaluates the model's performance on unseen data to ensure that the model has good generalization ability and stability. Through the above technical solutions, a fully trained and validated digital twin model can be obtained, providing strong support for subsequent health management applications.

[0079] The effects of the above technical solution are as follows: A digital twin model built based on a user's initial and dynamic physiological data can accurately reflect each user's physiological characteristics and trends, providing a highly personalized foundation for subsequent health management applications. Through data cleaning and model training and validation, the accuracy and reliability of the digital twin model are effectively improved. The model can more accurately simulate and predict the user's physiological state and its dynamic changes, providing users with more scientific health management advice. The digital twin model can receive and process the user's dynamic physiological data in real time, dynamically adjusting health management advice according to changes in the user's physiological state. This dynamic adaptability makes health management services more aligned with the user's actual needs, improving the service's relevance and effectiveness. Health management applications based on digital twin models can provide strong decision support for professionals such as doctors and health managers. Through model simulation and prediction results, professionals can more accurately assess the user's health status, predict potential risks, and develop targeted health management plans. Personalized health management services and dynamically adaptive recommendations can significantly enhance the user experience. Users can obtain tailored health management plans according to their actual situation, better manage their health status, and improve their quality of life.

[0080] In one embodiment of the present invention, the structure of the digital twin model is as follows:

[0081] The input layer is used to input the initial physiological data and dynamic physiological data of the target user, and to perform unified and standardized processing on the initial physiological data and dynamic physiological data of the target user.

[0082] The multi-scale time window layer includes multiple parallel LSTM sub-networks, and each LSTM sub-network corresponds to a time window; wherein, the parallel LSTM sub-networks are used to process input sequences of different lengths to control the digital twin model to capture data patterns at different time scales.

[0083] An adaptive gated fusion layer is used to dynamically adjust the parameter values ​​corresponding to the importance and contribution of each LSTM sub-network based on the current input initial physiological data and dynamic physiological data, and to synthesize the output of the dynamically adjusted LSTM sub-networks to form a unified hidden layer.

[0084] The context-aware fusion layer combines the output of the adaptive gated fusion layer with contextual information from the target user (e.g., current activity type, time, etc.); this allows for dynamic adjustment of the model's response, making the output more personalized and accurate.

[0085] The output layer is used to output the diet and run type.

[0086] The working principle of the above technical solution is as follows: the input layer is responsible for receiving the initial physiological data and dynamic physiological data of the target user. The above data is the basis for model prediction, including static initial physiological data (such as age, gender, weight, etc.) and real-time changing dynamic physiological data (such as heart rate, blood pressure, exercise calories, etc.).

[0087] The input layer also performs standardized processing on the above data to ensure that data from different sources and of different magnitudes can be compared fairly in the model, thereby improving the model's prediction accuracy.

[0088] Multi-scale time window layer:

[0089] This layer contains multiple parallel LSTM (Long Short-Term Memory) subnetworks, each corresponding to a specific time window. LSTM is a special type of recurrent neural network (RNN) that can effectively handle long-term dependencies in sequential data.

[0090] By setting up multiple parallel LSTM subnetworks, the model can simultaneously capture data patterns at different time scales. This is crucial for understanding the long-term trends and short-term fluctuations in a user's physiological state.

[0091] Adaptive Gated Fusion Layer:

[0092] This layer dynamically adjusts the importance and contribution of each LSTM subnetwork based on the initial and dynamic physiological data input. This is achieved by learning a set of parameters that reflect the importance of information at different time scales to the current prediction task.

[0093] The adaptive gated fusion layer synthesizes the dynamically adjusted LSTM sub-network outputs to form a unified hidden layer representation. This representation integrates information from different time scales, providing a rich contextual basis for subsequent predictions.

[0094] Context-aware fusion layer:

[0095] This layer utilizes contextual information from the target user (such as current activity type, time, etc.) to further refine the model's response. This information is crucial for understanding the user's current physiological state and needs.

[0096] By combining contextual information with the output of the adaptive gated fusion layer, the model can generate more personalized and accurate predictions. This combination allows the model to flexibly adjust its prediction strategy based on the user's specific circumstances.

[0097] Output layer:

[0098] The output layer is responsible for generating the final diet and exercise type recommendations. These recommendations are based on the processing results of the preceding layers, taking into full account the user's initial physiological state, dynamic physiological changes, and contextual information.

[0099] The above technical solution achieves the following results: By comprehensively considering the user's initial physiological data, dynamic physiological data, and contextual information, the model can generate highly personalized recommendations for diet and exercise types. These recommendations are more closely aligned with the user's actual needs and lifestyle habits. Utilizing a multi-scale time window layer, the model can simultaneously capture data patterns at different time scales. This allows the model to more comprehensively understand the user's physiological state changes and short-term fluctuations, improving prediction accuracy and robustness. The adaptive gating fusion layer and context-aware fusion layer enable the model to dynamically adjust its prediction strategy based on the user's real-time input and contextual information. This adaptive adjustment capability allows the model to better adapt to complex and ever-changing realities, improving prediction flexibility and accuracy. Based on the above working principles and technical effects, this digital twin model can provide users with more accurate, personalized, and practical recommendations for diet and exercise types. This will help users better manage their health, improve their quality of life, and enhance the user experience.

[0100] The model described above employs a multi-scale time window layer, using LSTM subnetworks with varying time window sizes in parallel. This allows each subnetwork to focus on capturing data patterns over a specific time period. This design enables the model to simultaneously understand and integrate short-, medium-, and long-term data changes. For example, a short-time-window LSTM is more sensitive to rapid heart rate changes, while a long-time-window LSTM can identify long-term physiological trends or cyclical patterns. Through multi-scale time windows, the model can understand data from multiple dimensions, increasing its expressiveness and helping the model to more comprehensively understand complex physiological and behavioral patterns. This multi-level time series analysis is significant for predicting users' health status and behavior, providing more accurate health recommendations. In practical applications, users' physiological data is influenced by various factors, exhibiting different dynamics and complexities. The multi-scale time window layer enhances the model's adaptability to different data characteristics by processing sequences of varying lengths. For example, long-term patterns are more important during relatively calm daily activities, while short-term heart rate changes are more crucial during high-intensity exercise. Following the multi-scale time window layer, an adaptive gating mechanism intelligently fuses the outputs of each sub-network, dynamically adjusting the contribution weight of each time scale output based on the current input. This dynamic weight adjustment mechanism provides the model with more flexible data processing capabilities, enabling it to respond optimally to different types of inputs. The multi-scale time window layer allows the model to learn and predict at multiple time scales, thus better adapting to and handling complex and dynamically changing physiological data. The design of these layers not only improves model performance but also enhances its applicability and flexibility for different application scenarios.

[0101] In one embodiment of the present invention, the steps for obtaining the time window are as follows:

[0102] Step 1: Retrieve the preset base attenuation rate from the database;

[0103] Step 2: Monitor the frequency of changes in the collected dynamic physiological data in real time;

[0104] Step 3: Obtain the dynamic decay factor corresponding to each LSTM sub-network using the preset basic decay rate and the data change frequency of dynamic physiological data;

[0105] Step 4: Obtain the time window length corresponding to each LSTM sub-network using the dynamic decay factor; wherein, the time window length corresponding to each LSTM sub-network is obtained by the following formula:

[0106]

[0107] Among them, K iα(i, β, λ) represents the time window length corresponding to the i-th LSTM subnetwork; T represents the preset observation period length; α(i, β, λ) represents the dynamic decay factor corresponding to the i-th LSTM subnetwork; N represents the total number of LSTM subnetworks; w i α represents the preset initial weight value corresponding to the i-th LSTM subnetwork; α0 represents the preset base decay rate; β represents the frequency factor of dynamic physiological data change; i represents the index corresponding to each LSTM subnetwork; λ represents the preset periodic intensity value of dynamic physiological data.

[0108] The working principle of the above technical solution is as follows: First, a preset base attenuation rate is retrieved from the database. This attenuation rate is a basic parameter used for subsequent calculation of the dynamic attenuation factor, reflecting the general law of data change over time.

[0109] The frequency of changes in the collected dynamic physiological data is monitored in real time. This step is crucial for determining the time window, as the frequency of changes in dynamic physiological data directly reflects the activity level and trend of the user's physiological state. Using a preset baseline decay rate and the frequency of changes in dynamic physiological data, the dynamic decay factor corresponding to each LSTM subnetwork is calculated. This factor reflects the rate of data decay at the current time scale and is an important basis for determining the length of the time window.

[0110] Based on the dynamic decay factor, the time window length corresponding to each LSTM subnetwork is calculated using the formula described above. This formula considers the observation period, the dynamic decay factor, the total number of LSTM subnetworks, and the preset initial weight values ​​for each subnetwork, ensuring that the time window length reflects both the dynamic changes in the data and maintains the model's stability and predictive performance.

[0111] The effects of the above technical solution are as follows: By monitoring the frequency of changes in dynamic physiological data in real time and calculating the dynamic decay factor and time window length accordingly, the model can dynamically adapt to changes in the user's physiological state. This makes the model more flexible and accurate when processing data from different users at different time points. Assigning different time windows to each LSTM sub-network allows the model to simultaneously capture data patterns at different time scales. This enables a more comprehensive understanding of the user's physiological state, improving prediction accuracy and robustness. The dynamic adjustment of the time window considers the unique physiological characteristics of each user and the changing characteristics of dynamic physiological data. This allows the model to provide more personalized predictions and recommendations for each user, enhancing the user experience. By reasonably setting the length of the time window, the performance and stability of the LSTM sub-network can be optimized. This reduces the computational complexity of the model, improves prediction speed, and maintains high prediction accuracy.

[0112] By retrieving a preset baseline decay rate from the database, the above technical solution provides a basic framework for determining the time window. This preset value can be adjusted according to different application scenarios and physiological data, thereby increasing the flexibility of the solution. Real-time monitoring of the frequency of changes in the collected dynamic physiological data allows the determination of the time window to dynamically adapt to data changes. This ensures that the time window setting matches the actual situation of the data, improving the accuracy of the analysis. Simultaneously, by calculating the dynamic decay factor corresponding to each LSTM subnetwork using the preset baseline decay rate and the frequency of changes in the dynamic physiological data, the above technical solution considers the dynamic characteristics of the data and the characteristics of the LSTM subnetwork. The introduction of the dynamic decay factor allows the length of the time window to be dynamically adjusted as the data changes, helping to capture key information in the data. By calculating the time window length corresponding to each LSTM subnetwork using a formula, the above technical solution achieves precise control of the time window length. This precise calculation helps ensure that the changing trends and characteristics of the data can be accurately captured during the analysis process.

[0113] The above formula considers multiple factors, including the preset observation period, dynamic decay factor, total number of LSTM subnetworks, and preset initial weight values ​​for each LSTM subnetwork. The combined effect of these factors allows the time window length to better align with the actual data and analytical needs. By setting different time window lengths for different LSTM subnetworks, the above technical solution can more accurately capture the data features of interest to each subnetwork. This improves analytical efficiency and accuracy, making the analysis results more reliable.

[0114] Furthermore, the aforementioned technical solution also considers the dynamic characteristics of the data and the features of the LSTM subnetwork, thus better adapting to different types of data and analytical needs. By clearly defining the steps and calculation methods for obtaining the time window, the above technical solution enhances the interpretability of the model. This allows researchers to more clearly understand how the model processes data and makes predictions, thereby contributing to further improvement and optimization of the model.

[0115] In summary, the above technical solutions, by dynamically adjusting the length of the time window, enable the digital twin model to better adapt to changes in the user's physiological state, improving the accuracy and personalization of predictions, thereby providing users with more precise and practical health management advice. Furthermore, by introducing dynamic decay factors and accurately calculating the time window length, these solutions achieve flexible and precise control over the time window. Moreover, they improve analytical efficiency and accuracy, enhance model interpretability, and provide better support for different types of data and analytical needs.

[0116] One embodiment of the present invention utilizes the preset base decay rate and the data change frequency of dynamic physiological data to obtain the dynamic decay factor corresponding to each LSTM sub-network, including:

[0117] Step 301: Retrieve the frequency of changes in the collected dynamic physiological data;

[0118] Step 302: Obtain the frequency of change factor using the frequency of change in the dynamic physiological data; wherein, the frequency of change factor is obtained by the following formula:

[0119]

[0120] Where β represents the frequency factor of dynamic physiological data; r represents the adjustment parameter, and the value range of the adjustment parameter is [0.1, 0.9]; f 01 f 02 and f 03 These correspond to the percentage changes in heart rate, blood pressure, and calories burned during exercise, respectively, representing the maximum range of change. p01 v p02 and v p03 These correspond to the average rate of change of heart rate, blood pressure, and exercise calories consumed per unit time, respectively.

[0121] Step 303: Use the changing frequency factor in combination with the basic attenuation rate to obtain the dynamic attenuation factor corresponding to each LSTM sub-network.

[0122] The dynamic attenuation factor for each LSTM sub-network is obtained using the following formula:

[0123]

[0124] Where α(i, β, λ) represents the dynamic decay factor corresponding to the i-th LSTM subnetwork; α0 represents the preset base decay rate; β represents the frequency factor of dynamic physiological data change; i represents the index corresponding to each LSTM subnetwork, and i = 1, 2, 3, ..., N; N represents the total number of LSTM subnetworks; and λ represents the preset periodic intensity value of dynamic physiological data.

[0125] The working principle of the above technical solution is as follows: First, the frequency of change of the collected dynamic physiological data is retrieved. This frequency of change reflects the activity level of the user's physiological state at different time points. Using the frequency of change of the dynamic physiological data, combined with the formula mentioned above, a frequency factor is calculated. This formula considers the proportion of change corresponding to the maximum change amplitude of heart rate, blood pressure, and exercise calories burned, as well as their average rate of change per unit time. Simultaneously, an adjustment parameter is introduced to adjust the sensitivity of the frequency factor. Combining the frequency factor and the baseline decay rate, another formula is used to calculate the dynamic decay factor corresponding to each LSTM subnetwork. This formula considers the index of the LSTM subnetwork, the baseline decay rate, the frequency factor, and the preset periodic intensity value of the dynamic physiological data. Through this formula, a suitable dynamic decay factor can be assigned to each LSTM subnetwork for subsequent time window length calculations.

[0126] The effects of the above technical solution are as follows: By monitoring the frequency of changes in dynamic physiological data in real time and calculating the dynamic decay factor accordingly, the model can more flexibly adapt to changes in the user's physiological state. This allows the model to more accurately capture data trends when processing data from different users at different time points. Assigning different dynamic decay factors to each LSTM sub-network makes the model more accurate in capturing data patterns at different time scales. It enables a more comprehensive understanding of the user's physiological state, improving prediction accuracy and robustness. The calculation of the dynamic decay factor considers the unique physiological characteristics of each user and the changing characteristics of dynamic physiological data. This allows the model to provide more personalized predictions and recommendations for each user, better meeting their actual needs. By reasonably setting the dynamic decay factor, the performance and stability of the LSTM sub-network can be optimized. It can reduce the computational complexity of the model, improve prediction speed, and maintain high prediction accuracy.

[0127] In summary, the above technical solution, by dynamically adjusting the dynamic decay factor corresponding to each LSTM sub-network, enables the digital twin model to better adapt to changes in the user's physiological state, improves the accuracy and personalization of predictions, and thus provides users with more accurate and practical health management advice.

[0128] In one embodiment of the present invention, after the digital twin model is constructed, initial physiological data is collected from registered target users, and the dynamic physiological data of the target users is dynamically adjusted and collected according to the dynamic changes in the dynamic physiological data of the target users, thereby obtaining the initial physiological data and dynamic physiological data of the registered target users, including:

[0129] S201. After the digital twin model is constructed, an initial physiological data input instruction is sent to the registered target user, and the initial physiological data input by the registered target user is received in real time; wherein, the initial physiological data includes age, gender, weight, height and basal metabolism.

[0130] S202. Real-time collection of dynamic physiological data of the registered target user through wearable devices worn by the registered target user; wherein, the dynamic physiological data includes heart rate, blood pressure and exercise calories;

[0131] S203. Extract the frequency of change of the dynamic physiological data of the registered target users and dynamically adjust the collection frequency of the dynamic physiological data.

[0132] The working principle of the above technical solution is as follows: After the digital twin model is constructed, the system first sends an initial physiological data input instruction to the registered target user. This instruction guides the user to input their basic physiological information, such as age, gender, weight, height, and basal metabolic rate. This information forms the foundational static data for building the user's digital twin model. After the user inputs this data through the system interface or designated channels, the system receives and stores this initial physiological data in real time. The system then collects dynamic physiological data in real time through wearable devices worn by the registered user. These wearable devices continuously monitor key physiological indicators such as the user's heart rate, blood pressure, and calorie expenditure, and transmit this data to the system in real time. During the data collection process, the system extracts the frequency of change in the dynamic physiological data, i.e., how quickly the aforementioned physiological indicators change over time. Based on the frequency of change, the system dynamically adjusts the collection frequency to ensure that data is collected more frequently when changes are rapid, in order to capture more detailed trends; while when data changes are slow, the collection frequency can be appropriately reduced to save resources.

[0133] The above technical solution achieves the following results: By collecting users' initial and dynamic physiological data, the system can construct a highly personalized digital twin model. This model can more accurately reflect the user's physiological state and trends, providing a solid foundation for subsequent health management and prediction. Real-time collection of dynamic physiological data, with the collection frequency dynamically adjusted based on data change frequency, ensures the system can capture subtle changes in the user's physiological state, improving prediction accuracy.

[0134] Wearable devices collect data in real time, enabling the system to instantly obtain the user's latest physiological status information. This real-time capability provides the system with dynamic response capabilities, allowing it to issue timely warnings or suggestions when abnormalities occur in the user's physiological state. The mechanism of dynamically adjusting the collection frequency further enhances the system's real-time response capability, ensuring sufficient data to support decision-making at critical moments. By dynamically adjusting the collection frequency based on data change frequency, the system can optimize resource usage while ensuring data accuracy. Reducing the collection frequency when data changes slowly reduces the load on data transmission and storage, lowering system energy consumption and costs. Users only need to wear the wearable device and input a small amount of initial physiological data to enjoy personalized, real-time health management services. This convenient operation enhances the user experience, making users more willing to participate in the health management process.

[0135] In summary, the above-mentioned technical solution provides strong support for the application of digital twin models in the field of health management through an efficient and accurate physiological data collection mechanism. It not only improves the accuracy and real-time performance of predictions but also optimizes resource utilization and enhances the user experience.

[0136] In one embodiment of the present invention, dynamically adjusting the collection frequency of dynamic physiological data by extracting the change frequency of the dynamic physiological data of the registered target user includes:

[0137] S2031. Extract the frequency of changes in heart rate, blood pressure, and exercise calories consumed from the dynamic physiological data of the registered target user.

[0138] S2032. Obtain the change frequency parameters corresponding to the heart rate, blood pressure, and exercise calories burned using the change frequency of the heart rate, blood pressure, and exercise calories burned;

[0139] S2033. The acquisition frequency of the dynamic physiological data is adjusted using the frequency parameters corresponding to the changes in heart rate, blood pressure, and exercise calories burned. The adjusted acquisition frequency of the dynamic physiological data is obtained using the following formula:

[0140]

[0141] Among them, f new Indicates the adjusted frequency of dynamic physiological data acquisition; f y ξ represents the frequency of dynamic physiological data collection before adjustment; ξ represents the preset adjustment coefficient, with a value range of 0.26-0.41; X 01 X represents the frequency parameter corresponding to the change in heart rate; 02 The parameter representing the frequency of change in blood pressure; X 03 This represents the frequency parameter corresponding to the change in calories burned during exercise.

[0142] The working principle of the above technical solution is as follows: The system first extracts the dynamic physiological data of registered target users, which includes key physiological indicators such as heart rate, blood pressure, and exercise calories burned. Next, the system analyzes the time-series data of these physiological indicators and calculates their respective frequencies of change. These frequencies of change reflect how quickly the physiological indicators change over time.

[0143] The system utilizes the extracted frequency changes of heart rate, blood pressure, and exercise calories burned to calculate corresponding frequency parameters using a specific algorithm or formula. These parameters represent further processing or quantification of the original frequency changes, used for subsequent acquisition frequency adjustments. Based on the calculated frequency parameters corresponding to heart rate, blood pressure, and exercise calories burned, and a preset adjustment coefficient (such as ξ), the system calculates the adjusted dynamic physiological data acquisition frequency (fnew) using the aforementioned formula. This formula comprehensively considers the original acquisition frequency (fy), the frequency parameters of various physiological indicators (X01, X02, X03), and the preset adjustment coefficient (ξ), ensuring that the acquisition frequency can adaptively adjust according to the dynamic changes in physiological data.

[0144] The effects of the above technical solution are as follows: By dynamically adjusting the acquisition frequency, the system can acquire data more frequently when physiological data changes rapidly, thereby more accurately capturing the changing trends of physiological state. This improves the accuracy of subsequent health predictions or diagnoses. When physiological data changes slowly, the system appropriately reduces the acquisition frequency, reducing the load on data transmission and storage, thus saving resources (such as power and network bandwidth). This dynamic adjustment mechanism helps extend the lifespan of wearable devices and reduce overall operating costs.

[0145] For users, dynamically adjusting the data collection frequency means that in most cases, the system will not disturb them too frequently (e.g., through vibration or reminders). Simultaneously, when needed (e.g., when physiological states are abnormal), the system can collect data promptly and issue warnings or suggestions, thereby improving user experience and satisfaction. The above technical solution provides a flexible data collection mechanism for the application of digital twin models in the field of health management. It allows the system to dynamically adjust its collection strategy based on the user's actual physiological state, thus better adapting to the needs of different users and in different scenarios.

[0146] In summary, the above technical solution improves both the accuracy and efficiency of data acquisition by extracting the frequency of change in dynamic physiological data and adjusting the acquisition frequency accordingly, thus possessing significant practical application value.

[0147] One embodiment of the present invention utilizes the frequency of change of the heart rate, blood pressure, and exercise calories burned to obtain the corresponding frequency parameters of change of the heart rate, blood pressure, and exercise calories burned, including:

[0148] Step 1: Obtain the heart rate change frequency parameter corresponding to the heart rate using the heart rate change frequency contained in the dynamic physiological data; wherein, the heart rate change frequency parameter is obtained by the following formula:

[0149]

[0150] Among them, X 01 X represents the frequency parameter corresponding to the change in heart rate; b01 This represents the standard deviation of heart rate; m 01 Indicates the number of heart rate measurements; H i H represents the heart rate value corresponding to the i-th data collection; i+1 H represents the heart rate value corresponding to the (i+1)th data collection; f This indicates the maximum permissible fluctuation range of the heart rate value;

[0151] Step 2: Obtain the blood pressure change frequency parameter corresponding to the blood pressure using the blood pressure change frequency contained in the dynamic physiological data; wherein, the blood pressure change frequency parameter is obtained by the following formula:

[0152]

[0153] Among them, X 02 The parameter representing the frequency of change in blood pressure; X x02 X represents the standard deviation of diastolic blood pressure; y02 This represents the standard deviation of systolic blood pressure; m 02 Indicates the number of blood pressure measurements; S 01i+1 and S 01i S represents the diastolic blood pressure values ​​corresponding to the (i+1)th and ithth measurements; 02i+1 and S 02i This represents the systolic blood pressure values ​​collected in the (i+1)th and ithth measurements.

[0154] Step 3: Obtain the frequency parameter corresponding to the change in exercise-induced calorie expenditure using the frequency of change in the dynamic physiological data; wherein, the frequency parameter corresponding to the change in exercise-induced calorie expenditure is obtained by the following formula:

[0155]

[0156] Among them, X 03 X represents the frequency parameter corresponding to the change in calories burned during exercise; b03 The standard deviation of calories burned during exercise; m 03 E represents the number of times calories burned during exercise. i E represents the calorie expenditure value corresponding to the i-th data collection. i+1E represents the calorie expenditure value corresponding to the (i+1)th data collection. f This indicates the preset reference value for the fluctuation of exercise calories.

[0157] The working principle of the above technical solution is as follows: First, the system extracts heart rate data from dynamic physiological data and calculates the standard deviation of heart rate (X). b01 This is used to measure the dispersion of heart rate data. Next, the system uses the number of heart rate data collections (m) to measure the dispersion of the heart rate data. 01 ) and heart rate values ​​collected in two consecutive intervals (H i and H i+1 The heart rate change frequency parameter (X) is calculated using the above formula. 01 The above formula takes into account the overall dispersion of heart rate data and the variation between adjacent data points, thus reflecting the frequency characteristics of heart rate changes.

[0158] The system extracts blood pressure data from dynamic physiological data, including diastolic and systolic blood pressure, and calculates their standard deviations (X). x02 and X y02 Then, the system uses the number of times blood pressure data is collected (m) 02 ) and the diastolic and systolic blood pressure values ​​from two consecutive measurements (S 01i S 01i+1 S 02i and S 02i+1 The frequency parameter (X) corresponding to the change in blood pressure is calculated using the above formula. 02 The above formula comprehensively considers the changing characteristics of diastolic and systolic blood pressure, as well as the changes between adjacent data points, thus more accurately reflecting the frequency of blood pressure changes. The system extracts exercise-related calorie expenditure data from dynamic physiological data and calculates its standard deviation (X). b03 ).

[0159] Next, the system utilizes the number of times exercise calorie consumption data is collected (m) 03 ) and the energy expenditure values ​​collected in two consecutive measurements (E) i and E i+1 The frequency parameter (X) corresponding to the change in exercise calories is calculated using the above formula. 03 The above formula takes into account the overall dispersion of exercise calorie expenditure data and the changes between adjacent data points, thus reflecting the frequency characteristics of changes in exercise calorie expenditure.

[0160] The effects of the above technical solution are as follows: By calculating the frequency parameters corresponding to changes in heart rate, blood pressure, and exercise calories burned, the system can more accurately reflect changes in the user's physiological state. These parameters comprehensively consider the overall dispersion of physiological data and the changes between adjacent data points, thus providing more accurate and comprehensive physiological state information. The calculation of the frequency parameter provides an important basis for the system to dynamically adjust the collection frequency of physiological data. When physiological states change rapidly, the system can increase the collection frequency to capture more physiological information; when physiological states change slowly, the system can appropriately reduce the collection frequency to save resources. Simultaneously, accurate calculation of the frequency parameter helps improve the accuracy and efficiency of health management. The system can provide personalized health management suggestions or warnings based on the user's actual physiological state, thereby better protecting the user's health. By dynamically adjusting the collection frequency and providing accurate physiological state information, the system can optimize the user experience. Users can experience more personalized and accurate health management services, thereby increasing their satisfaction and trust in the system.

[0161] On the other hand, by calculating the frequency parameters of change in heart rate, blood pressure, and calorie expenditure, the system can quickly identify significant changes in a user's physiological state. This helps the system respond more quickly, such as issuing timely alerts or adjusting health management strategies when abnormal physiological states are detected. This real-time capability is crucial for monitoring and intervening in emergency health conditions. The introduction of frequency parameters provides a new dimension to the analysis of physiological data. In addition to traditional statistical indicators such as mean and standard deviation, the system can also use these parameters to analyze the dynamic changes in a user's physiological state in depth. This helps to discover potential physiological problems or trends, providing more comprehensive and in-depth insights for health management. Simultaneously, by accurately calculating the frequency parameters of change in a user's heart rate, blood pressure, and calorie expenditure, the system can more accurately understand the user's physiological characteristics and needs. This helps the system provide more personalized health management services, such as customizing exercise plans and dietary recommendations based on the user's physiological state. This personalized service can better meet user needs and improve the effectiveness of health management.

[0162] By dynamically adjusting the frequency of physiological data collection, the system can rationally allocate resources based on the user's actual physiological state. When the physiological state is stable, reducing the collection frequency minimizes unnecessary resource consumption; conversely, when the physiological state fluctuates significantly, increasing the collection frequency ensures sufficient data to support accurate health analysis and decision-making. This resource allocation strategy helps reduce system operating costs and improve overall efficiency. When users see accurate health advice and warnings provided by the system based on changes in their physiological state, they become more attentive to their health. This increased attention helps enhance users' health awareness, encouraging them to participate more actively in health management. Active user participation not only improves the effectiveness of health management but also strengthens their trust in and satisfaction with the system.

[0163] In summary, the above technical solution, by calculating the frequency parameters corresponding to changes in heart rate, blood pressure, and calorie expenditure, achieves the function of accurately capturing changes in the user's physiological state and dynamically adjusting the acquisition frequency, which has significant practical application value. This technical solution, by calculating the frequency parameters corresponding to changes in heart rate, blood pressure, and calorie expenditure, not only improves the accuracy and efficiency of health management but also brings about multiple technical benefits, including real-time performance, personalized services, resource optimization, and enhanced user health awareness. These effects collectively promote the improvement of users' health status and the enhancement of health management levels.

[0164] In one embodiment of the present invention, a digital twin model is used to simulate and predict the initial and dynamic physiological data of the registered target user, obtain a diet and exercise plan that matches the diet and exercise type of the registered target user, and recommend it to the registered target user, including:

[0165] S301. Use a digital twin model to simulate and predict the initial and dynamic physiological data of the registered target user, and obtain the diet and exercise type that match the initial and dynamic physiological data of the registered target user.

[0166] S302. Based on the diet and exercise types that match the initial and dynamic physiological data of the registered target users, retrieve diet and exercise plans from the database and recommend them to the registered target users.

[0167] The working principle of the above technical solution is as follows: First, the system collects and integrates the initial physiological data (such as age, gender, weight, height, and basal metabolic rate) and dynamic physiological data (such as heart rate, blood pressure, and exercise calories burned) of registered target users. Then, this physiological data is input into a pre-built digital twin model. The digital twin model uses advanced algorithms and machine learning techniques to simulate and predict the user's physiological state. By simulating physiological responses and trends under different physiological states, the model can assess the potential effects of different diets and exercise programs on the user. Based on the simulation and prediction results, the system can identify the diet and exercise types that best match the user's current physiological state. For example, for users who need to lose weight, the model will recommend a low-fat, high-protein diet and a moderate-intensity aerobic exercise program.

[0168] Once the system identifies a diet and exercise type that matches the user's physiological data, it retrieves the corresponding diet and exercise plan from a pre-set database. This plan includes specific dietary plans (such as daily meal recipes and food intake) and exercise plans (such as exercise type, duration, and intensity). Finally, the system recommends the retrieved personalized diet and exercise plan to registered target users. The recommendations are presented in text, charts, or application interfaces for easy understanding and implementation by the user.

[0169] The effects of the above technical solution are as follows: Through the simulation and prediction capabilities of the digital twin model, the system can provide highly personalized diet and exercise program recommendations for each user. These personalized recommendations help users manage their health more accurately, improving their quality of life and life expectancy. The digital twin model can simulate and predict the user's physiological state changes in real time, thus providing a scientific basis for the formulation of diet and exercise programs. This accurate prediction helps optimize the effectiveness of the program, reducing unnecessary trial and error and adjustments. Personalized diet and exercise program recommendations can significantly improve user experience and satisfaction. Users no longer need to blindly try various diet and exercise programs, but can directly obtain the most suitable health management advice for themselves.

[0170] Through continuous personalized recommendations and feedback mechanisms, the system helps users gradually develop healthy lifestyle habits. These changes in health behaviors are of great significance for the prevention and control of chronic diseases.

[0171] In summary, the above technical solutions utilize the simulation and prediction capabilities of digital twin models to provide registered target users with personalized diet and exercise program recommendations, thereby achieving precise, personalized, and intelligent health management.

[0172] This invention proposes a diet and exercise recommendation system based on digital twins, such as... Figure 2As shown, the diet and exercise recommendation system based on digital twins includes:

[0173] The digital twin model building module is used to build digital twin models using the initial and dynamic physiological data of users that have been collected.

[0174] The data acquisition module is used to collect initial physiological data of registered target users after the digital twin model is built, and to dynamically adjust the collection of dynamic physiological data of target users according to the dynamic changes of dynamic physiological data of target users, so as to obtain the initial physiological data and dynamic physiological data of registered target users.

[0175] The program recommendation module is used to simulate and predict the initial and dynamic physiological data of the registered target users using a digital twin model, obtain diet and exercise programs that match the diet and exercise types of the registered target users, and recommend them to the registered target users.

[0176] The working principle of the above technical solution is as follows: Historical physiological data of users is retrieved from the database, including static initial physiological data (such as age, gender, weight, height, and basal metabolic rate) and dynamic physiological data (such as heart rate, blood pressure, and exercise calories burned). This data forms the foundation for building the digital twin model. Cleaning the retrieved data is a crucial data quality control process. By removing invalid data (such as data with too many missing values ​​or data that is clearly contrary to common sense), duplicate data, and erroneous data, the accuracy and reliability of subsequent model construction are ensured. The preprocessed data is used to build the digital twin model. The completed digital twin model is then trained and validated. The training process involves adjusting the model parameters using a large amount of data to enable the model to more accurately reflect actual conditions; the validation process evaluates the model's performance on unseen data to ensure the model's generalization ability and stability.

[0177] The above technical solution achieves the following results: A digital twin model, constructed based on the user's initial and dynamic physiological data, accurately reflects each user's physiological characteristics and trends, providing a highly personalized foundation for subsequent diet and exercise recommendations. Through data cleaning and model training validation, the accuracy and reliability of the digital twin model are effectively improved, enabling it to more accurately predict the user's physiological responses under different diet and exercise programs. The model can receive and process the user's dynamic physiological data in real time, dynamically adjusting diet and exercise recommendations based on changes in the user's physiological state, ensuring the scientific validity and timeliness of the suggestions. The diet and exercise recommendation method based on the digital twin model can provide users with more precise and effective health management services, helping them better manage their health, prevent disease, and improve their quality of life.

[0178] In summary, the above technical solutions achieve precise management and effective intervention of users' health status by constructing highly personalized digital twin models and combining them with dynamic physiological data to update recommendation schemes in real time.

[0179] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for recommending diet and exercise based on digital twins, characterized in that, The diet and exercise recommendation method based on digital twins includes: Digital twin models are constructed using the initial and dynamic physiological data of users that have been collected; After the digital twin model is built, initial physiological data is collected from the registered target users, and the dynamic physiological data of the target users is dynamically adjusted and collected according to the dynamic changes of the target users' dynamic physiological data to obtain the initial physiological data and dynamic physiological data of the registered target users. The initial and dynamic physiological data of the registered target users are simulated and predicted using a digital twin model to obtain diet and exercise plans that match the diet and exercise type of the registered target users, and then recommended to the registered target users. The structure of the digital twin model is as follows: The input layer is used to input the initial physiological data and dynamic physiological data of the target user, and to perform unified and standardized processing on the initial physiological data and dynamic physiological data of the target user. The multi-scale time window layer includes multiple parallel LSTM sub-networks, and each LSTM sub-network corresponds to a time window; wherein, the parallel LSTM sub-networks are used to process input sequences of different lengths to control the digital twin model to capture data patterns at different time scales. An adaptive gated fusion layer is used to dynamically adjust the parameter values ​​corresponding to the importance and contribution of each LSTM sub-network based on the current input initial physiological data and dynamic physiological data, and to synthesize the output of the dynamically adjusted LSTM sub-networks to form a unified hidden layer. The context-aware fusion layer combines the output of the adaptive gated fusion layer with contextual information from the target user; this allows for dynamic adjustment of the model's response, making the output more personalized and accurate. The output layer is used to output the diet and run type; The steps for obtaining the time window are as follows: Retrieve preset baseline decay rates from the database; monitor the frequency of data changes in collected dynamic physiological data in real time; The dynamic attenuation factor corresponding to each LSTM subnetwork is obtained by using the preset basic attenuation rate and the data change frequency of dynamic physiological data. The time window length corresponding to each LSTM sub-network is obtained using the dynamic decay factor; wherein, the time window length corresponding to each LSTM sub-network is obtained by the following formula: in, K i Indicates the first i The time length of the time window corresponding to each LSTM subnetwork; T Indicates the preset observation period duration; α ( i , β , λ ) indicates the first i The dynamic decay factor corresponding to each LSTM subnetwork; N This represents the total number of LSTM subnetworks; w i Indicates the first i Each LSTM subnetwork corresponds to a preset initial weight value; α 0 indicates the preset base decay rate; β A frequency factor representing the change in dynamic physiological data; i This represents the index corresponding to each LSTM subnetwork; λ This represents the periodic intensity value of the preset dynamic physiological data.

2. The diet and exercise recommendation method based on digital twins according to claim 1, characterized in that, Digital twin models are constructed using the initial and dynamic physiological data of users that have been collected, including: The system retrieves initial and dynamic physiological data of users from the database. The initial physiological data includes age, gender, weight, height, and basal metabolic rate. The dynamic physiological data includes heart rate, blood pressure, and calories burned during exercise. The collected initial and dynamic physiological data of users are cleaned to remove invalid, duplicate, and erroneous data, and preprocessed initial and dynamic physiological data are obtained. A digital twin model is constructed using the preprocessed initial physiological data and dynamic physiological data. The constructed digital twin model is then trained and validated to obtain the completed digital twin model.

3. The diet and exercise recommendation method based on digital twins according to claim 1, characterized in that, The dynamic decay factor for each LSTM subnetwork is obtained using the preset base decay rate and the data change frequency of dynamic physiological data, including: The frequency of data changes in the collected dynamic physiological data; The frequency of change factor is obtained using the frequency of change in the dynamic physiological data; wherein, the frequency of change factor is obtained by the following formula: in, β A frequency factor representing the change in dynamic physiological data; r This represents an adjustment parameter, and the value range of the adjustment parameter is [0.1, 0.9]; f 01 , f 02 and f 03 These correspond to the percentage changes corresponding to the maximum range of change in heart rate, blood pressure, and exercise calories burned, respectively. v p01 , v p02 and v p03 These correspond to the average rate of change of heart rate, blood pressure, and exercise calories consumed per unit time, respectively. The dynamic attenuation factor corresponding to each LSTM sub-network is obtained by combining the changing frequency factor with the basic attenuation rate. The dynamic attenuation factor for each LSTM sub-network is obtained using the following formula: in, α ( i , β , λ ) indicates the first i The dynamic decay factor corresponding to each LSTM subnetwork; α 0 indicates the preset base decay rate; β A frequency factor representing the change in dynamic physiological data; i This represents the index corresponding to each LSTM subnetwork, and... i =1,2,3,... N ; N This represents the total number of LSTM subnetworks; λ This represents the periodic intensity value of the preset dynamic physiological data.

4. The diet and exercise recommendation method based on digital twins according to claim 1, characterized in that, After the digital twin model is built, initial physiological data is collected from registered target users. The dynamic physiological data of the target users is then dynamically adjusted based on their dynamic changes, resulting in the acquisition of both initial and dynamic physiological data for the registered target users. After the digital twin model is constructed, an initial physiological data input instruction is sent to the registered target user, and the initial physiological data input by the registered target user is received in real time; wherein, the initial physiological data includes age, gender, weight, height and basal metabolic rate; The system collects real-time dynamic physiological data of registered target users through wearable devices; wherein the dynamic physiological data includes heart rate, blood pressure, and calories burned during exercise. The frequency of dynamic physiological data collection is dynamically adjusted by extracting the frequency of change of the dynamic physiological data of the registered target users.

5. The diet and exercise recommendation method based on digital twins according to claim 4, characterized in that, The frequency of dynamic physiological data collection is dynamically adjusted by extracting the frequency of change of the dynamic physiological data of the registered target users, including: Extract the frequency of changes in heart rate, blood pressure, and exercise calories burned from the dynamic physiological data of the registered target users; The frequency parameters corresponding to the changes in heart rate, blood pressure, and exercise calories are obtained using the frequency of change of these parameters. The acquisition frequency of the dynamic physiological data is adjusted using the frequency parameters corresponding to the changes in heart rate, blood pressure, and exercise calories burned. The adjusted acquisition frequency of the dynamic physiological data is obtained using the following formula: in, f new This indicates the adjusted frequency of dynamic physiological data collection; f y This indicates the frequency of dynamic physiological data collection before adjustment; ξ This represents the preset adjustment coefficient, with a value range of 0.26-0.41; X 01 This parameter represents the frequency of change in heart rate. X 02 A parameter representing the frequency of change in blood pressure; X 03 This represents the frequency parameter corresponding to the change in calories burned during exercise.

6. The diet and exercise recommendation method based on digital twins according to claim 5, characterized in that, Obtaining the frequency parameters corresponding to the changes in heart rate, blood pressure, and exercise calories burned using the frequency of these changes includes: The frequency of heart rate changes contained in the dynamic physiological data is used to obtain the frequency parameter corresponding to the heart rate; wherein, the frequency parameter corresponding to the heart rate is obtained by the following formula: in, X 01 This parameter represents the frequency of change in heart rate. X b01 This represents the standard deviation of the heart rate. m 01 Indicates the number of heart rate measurements taken; H i Indicates the first i The heart rate value corresponding to each collection; H i+1 Indicates the first i +1 heart rate value corresponding to the data collection; H f This indicates the maximum permissible fluctuation range of the heart rate value; The blood pressure change frequency parameter is obtained by utilizing the blood pressure change frequency contained in the dynamic physiological data; wherein, the blood pressure change frequency parameter is obtained by the following formula: in, X 02 A parameter representing the frequency of change in blood pressure; X x02 This represents the standard deviation corresponding to diastolic blood pressure; X y02 This represents the standard deviation of systolic blood pressure. m 02 Indicates the number of times blood pressure was collected; S 01i+1 and S 01i Indicates the first i +1st time and the first i The value corresponding to the diastolic blood pressure collected in the previous measurement; S 02i+1 and S 02i Indicates the first i +1st time and the first i The value corresponding to the systolic blood pressure collected in this instance; The frequency of change of exercise-induced calorie expenditure contained in the dynamic physiological data is used to obtain the frequency parameter corresponding to the exercise-induced calorie expenditure; wherein, the frequency parameter corresponding to the exercise-induced calorie expenditure is obtained by the following formula: in, X 03 The parameter representing the frequency of change of calories burned during exercise; X b03 This represents the standard deviation of calories burned during exercise. m 03 The number of times calories burned during exercise were collected; E i Indicates the first i The corresponding exercise calorie consumption value was collected each time; E i+1 Indicates the first i +1 collection of the corresponding exercise calorie consumption value; E f This indicates the preset reference value for the fluctuation of exercise calories.

7. The diet and exercise recommendation method based on digital twins according to claim 1, characterized in that, Using a digital twin model, the initial and dynamic physiological data of the registered target users are simulated and predicted to obtain diet and exercise plans that match the diet and exercise types of the registered target users, and recommendations are made to the registered target users, including: The initial and dynamic physiological data of the registered target users are simulated and predicted using a digital twin model to obtain diet and exercise types that match the initial and dynamic physiological data of the registered target users. Based on the diet and exercise types that match the initial and dynamic physiological data of the registered target users, diet and exercise plans are retrieved from the database and recommended to the registered target users.

8. A diet and exercise recommendation system based on digital twins, characterized in that, The digital twin-based diet and exercise recommendation system includes: The digital twin model building module is used to build digital twin models using the initial and dynamic physiological data of users that have been collected. The data acquisition module is used to collect initial physiological data of registered target users after the digital twin model is built, and to dynamically adjust the collection of dynamic physiological data of target users according to the dynamic changes of dynamic physiological data of target users, so as to obtain the initial physiological data and dynamic physiological data of registered target users. The program recommendation module is used to simulate and predict the initial and dynamic physiological data of the registered target users using a digital twin model, obtain diet and exercise programs that match the diet and exercise type of the registered target users, and recommend them to the registered target users. The structure of the digital twin model is as follows: The input layer is used to input the initial physiological data and dynamic physiological data of the target user, and to perform unified and standardized processing on the initial physiological data and dynamic physiological data of the target user. The multi-scale time window layer includes multiple parallel LSTM sub-networks, and each LSTM sub-network corresponds to a time window; wherein, the parallel LSTM sub-networks are used to process input sequences of different lengths to control the digital twin model to capture data patterns at different time scales. An adaptive gated fusion layer is used to dynamically adjust the parameter values ​​corresponding to the importance and contribution of each LSTM sub-network based on the current input initial physiological data and dynamic physiological data, and to synthesize the output of the dynamically adjusted LSTM sub-networks to form a unified hidden layer. The context-aware fusion layer combines the output of the adaptive gated fusion layer with contextual information from the target user; this allows for dynamic adjustment of the model's response, making the output more personalized and accurate. The output layer is used to output the diet and run type; The steps for obtaining the time window are as follows: Retrieve preset baseline decay rates from the database; monitor the frequency of data changes in collected dynamic physiological data in real time; The dynamic attenuation factor corresponding to each LSTM subnetwork is obtained by using the preset basic attenuation rate and the data change frequency of dynamic physiological data. The time window length corresponding to each LSTM sub-network is obtained using the dynamic decay factor; wherein, the time window length corresponding to each LSTM sub-network is obtained by the following formula: in, K i Indicates the first i The time length of the time window corresponding to each LSTM subnetwork; T Indicates the preset observation period duration; α ( i , β , λ ) indicates the first i The dynamic decay factor corresponding to each LSTM subnetwork; N This represents the total number of LSTM subnetworks; w i Indicates the first i Each LSTM subnetwork corresponds to a preset initial weight value; α 0 indicates the preset base decay rate; β A frequency factor representing the change in dynamic physiological data; i This represents the index corresponding to each LSTM subnetwork; λ This represents the periodic intensity value of the preset dynamic physiological data.

Images