Artificial intelligence, ai, data analysis method and apparatus based on wearable devices

Abstract

Embodiments of the present application disclose an artificial intelligence (AI) data analysis method and device based on a wearable device. The method comprises: collecting multi-source heterogeneous data of a target user, processing the multi-source heterogeneous data by an AI model to extract hidden variables, including at least one of the following: performing data standardization processing on the multi-source heterogeneous data to obtain a three-dimensional data matrix; wherein the three-dimensional data matrix comprises a first matrix and a second matrix; extracting time sequence features from the first matrix based on a time correlation algorithm; extracting non-time sequence features from the second matrix based on hot encoding or embedding layer conversion; fusing the time sequence features and the non-time sequence features based on the synergistic effect of dynamic physiological processes and static physiological states to obtain hidden variables; obtaining a metabolic portrait of the target user based on the hidden variables; generating an intervention strategy based on the metabolic portrait of the target user and fat gain / loss information; and performing an intervention operation according to the intervention strategy.

Description

Technical Field

[0001] This invention relates to the field of big health technology, and in particular to an artificial intelligence (AI) intervention method and device based on time-series data analysis of wearable devices. Background Technology

[0002] With the widespread adoption of wearable devices such as smart bracelets, body fat scales, and blood glucose meters, existing technologies have emerged that utilize these devices to collect user data and combine it with artificial intelligence for weight intervention. This typically involves collecting data such as heart rate, weight, and blood glucose, processing it through models to generate simple analytical results, and then providing dietary or exercise suggestions to assist in weight management. However, there are several limitations in practical applications:

[0003] Data collected by wearable devices is mostly isolated records or simple summaries, which are difficult to reflect dynamic physiological processes related to weight, such as the synergistic changes in nighttime water metabolism and energy consumption.

[0004] When the model processes data, it extracts more basic statistical features, which leads to a lag in monitoring weight changes;

[0005] The generated user metabolism profiles often misjudge weight fluctuations, such as misidentifying short-term edema as fat gain.

[0006] Intervention strategies often employ general templates. Summary of the Invention

[0007] In view of this, embodiments of the present invention provide an artificial intelligence (AI) data analysis method and apparatus based on wearable devices. The technical solution of the present invention is implemented as follows:

[0008] The first aspect of this disclosure provides an AI data analysis method based on wearable devices, comprising: collecting multi-source heterogeneous data from a target user, wherein the multi-source heterogeneous data includes at least data collected through a wearable device; processing the multi-source heterogeneous data using an AI model to extract latent variables; wherein the step of processing the multi-source heterogeneous data using an AI model to extract latent variables includes at least one of the following: performing data standardization on the multi-source heterogeneous data to obtain a three-dimensional data matrix; wherein the three-dimensional data matrix includes a first matrix and a second matrix; the first matrix includes time-series data extracted from the multi-source heterogeneous data; the second matrix includes non-time-series data extracted from the multi-source heterogeneous data; and performing data analysis based on a time correlation algorithm. Temporal features are extracted from the first matrix; the temporal features include at least one of local temporal features and global temporal features; non-temporal features are extracted from the second matrix based on hot coding or embedding layer transformation; the hidden variables are obtained by fusing the temporal features and the non-temporal features based on the synergistic effect of dynamic physiological processes and static physiological states; the temporal features are used to reflect the dynamic physiological processes of the target user; the non-temporal features are used to reflect the static physiological states of the target user; a metabolic profile of the target user is obtained based on the hidden variables; the metabolic profile is used to reflect the water metabolism and energy metabolism of the target user; an intervention strategy is generated based on the metabolic profile of the target user and fat gain / loss information; and the intervention operation is executed according to the intervention strategy.

[0009] A second aspect of this disclosure provides an artificial intelligence (AI) data analysis device based on a wearable device, comprising:

[0010] A data acquisition module is used to acquire multi-source heterogeneous data from a target user, wherein the multi-source heterogeneous data includes at least data acquired through a wearable device; an extraction module is used to process the multi-source heterogeneous data using an AI model to extract hidden variables; wherein the extraction module is specifically used to perform at least one of the following: performing data standardization processing on the multi-source heterogeneous data to obtain a three-dimensional data matrix; wherein the three-dimensional data matrix includes a first matrix and a second matrix; the first matrix includes time-series data extracted from the multi-source heterogeneous data; the second matrix includes non-time-series data extracted from the multi-source heterogeneous data; and extracting hidden variables from the multi-source heterogeneous data based on a time correlation algorithm. A temporal feature is extracted from a matrix; the temporal feature includes at least one of local temporal features and global temporal features; non-temporal features are extracted from the second matrix based on hot coding or embedding layer transformation; based on the synergistic effect of dynamic physiological processes and static physiological states, the temporal feature and the non-temporal feature are fused to obtain the hidden variable; the temporal feature is used to reflect the dynamic physiological processes of the target user; the non-temporal feature is used to reflect the static physiological state of the target user; an acquisition module is used to acquire a metabolic profile of the target user based on the hidden variable; the metabolic profile is used to reflect the water metabolism and energy metabolism of the target user;

[0011] The strategy module is used to generate intervention strategies based on the target user's metabolic profile and fat gain / loss information;

[0012] An intervention module is used to perform intervention operations according to the intervention strategy.

[0013] A third aspect of this disclosure provides a computer-readable medium having instructions stored thereon that, when executed by one or more processors, cause the processors to perform the AI ​​data analysis method based on a wearable device as described in any of the foregoing technical solutions.

[0014] The technical solution provided in this disclosure overcomes the limitations of a single data source by collecting multi-source heterogeneous data (including wearable device data) from target users. It simultaneously covers multi-dimensional information such as user physiology and behavior, providing more complete and dynamic basic data support for subsequent analysis. A three-dimensional data matrix is ​​used to standardize the multi-source heterogeneous data, classifying and integrating time-series and non-time-series data. This solves the processing difficulties caused by differences in heterogeneous data formats and lays a structured foundation for feature extraction. For time-series data, local and global time-series features are extracted, accurately capturing the instantaneous features and long-term trends of the user's dynamic physiological processes. For non-time-series data, features are extracted through hot encoding or embedding layer transformation, effectively characterizing the inherent attributes of the user's static physiological state. Based on the synergistic effect of dynamic physiological processes and static physiological states, the two types of features are fused, overcoming the limitations of a single feature dimension. This allows the extracted hidden variables to more comprehensively and accurately reflect the internal correlation and overall pattern of the user's physiological state, avoiding analytical bias caused by feature fragmentation. Metabolic profiles generated based on hidden variables focus on water and energy metabolism, accurately mapping the core metabolic state of target users. Compared to traditional broad physiological assessments, this approach is more aligned with individual metabolic characteristics, providing a clear target basis for subsequent interventions. Combining metabolic profiles with fat gain / loss information to generate intervention strategies enables multi-dimensional correlation analysis of physiological state, metabolic characteristics, and fat changes. This allows intervention strategies to match users' current metabolic needs while specifically addressing fat gain / loss issues, avoiding indiscriminate intervention. Furthermore, strategy-based intervention execution ensures the feasibility and sustainability of the intervention's effects. In summary, this method significantly enhances the AI ​​health analysis capabilities of wearable devices through multi-source data integration, precise feature extraction, personalized profile construction, and closed-loop intervention, more efficiently supporting personalized health management, metabolic regulation, and body fat control scenarios. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0016] Figure 1 A flowchart illustrating an artificial intelligence (AI) data analysis method based on wearable devices, provided as an embodiment of the present invention;

[0017] Figure 2 A flowchart illustrating another AI data analysis method based on wearable devices provided in an embodiment of the present invention;

[0018] Figure 3 This is a schematic diagram of the structure of an artificial intelligence (AI) data analysis device based on a wearable device, provided in an embodiment of the present invention.

[0019] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0021] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] like Figure 1 As shown, this disclosure provides an AI data analysis method based on wearable devices, including:

[0023] S1110: Collect multi-source heterogeneous data of the target user, wherein the multi-source heterogeneous data includes at least data collected through wearable devices;

[0024] S1120: Process the multi-source heterogeneous data using an AI model to extract hidden variables;

[0025] S1130: Based on the hidden variables, obtain the metabolic profile of the target user; the metabolic profile is used to reflect the target user's water metabolism and energy metabolism;

[0026] S1140: Generate an intervention strategy based on the target user's metabolic profile and fat gain / loss information;

[0027] S1150: Perform the intervention operation according to the intervention strategy.

[0028] In some embodiments, this wearable device-based AI data analysis method can be used in various electronic devices. These electronic devices can connect to big data service platforms such as health monitoring platforms and interact with the platform's servers.

[0029] For example, this AI data analysis method based on wearable devices can be used in one or more electronic devices. For example, the method can be used in mobile devices, which may be handheld devices and / or wearable devices programmed for the target user, specifically including but not limited to mobile phones, wristbands, ankle bracelets, etc. In other embodiments, this AI data analysis method based on wearable devices can also be used in medical assistive devices. For example, the medical assistive device may be a health status monitoring device for the target user in a hospital or clinic. The owner of the electronic device can be the target user themselves, or a relative, friend, or guardian of the target user.

[0030] The electronic device runs an application or app for health monitoring. In some embodiments, the application or app can monitor multiple health indicators for a health goal. For example, it can monitor the target user's bone health, muscle health, etc.

[0031] In some embodiments, multi-source heterogeneous data can be a collection of time-series data collected from different devices and platforms, which differ in type, format, and dimension, but are related to the user's metabolic state. Examples: smart bracelets (heart rate, steps, sleep stages), body fat scales (weight, body fat percentage), diet apps (calories, nutrients), smart water bottles (water intake, drinking time), etc., are integrated into a time-series dataset by timestamp, such as a user's heart rate fluctuations plus dietary calorie and body fat percentage changes from 6:00 to 22:00 daily from July 1st to July 7th.

[0032] In some embodiments, the AI ​​model may include a time-series data analysis model. For example, the AI ​​model may include, but is not limited to, machine learning models built on recurrent neural networks (RNNs), LSTMs, Transformers, etc., used to mine latent patterns and correlations in time-series data and extract abstract features reflecting metabolic states. Example: Using an LSTM model to process multi-source time-series data, the input is hourly heart rate plus daily dietary calories plus weekly body fat percentage, and the output is hidden variables, such as metabolic fluctuation index and water metabolism efficiency.

[0033] In some embodiments, hidden variables—potential parameters extracted by the AI ​​model from raw data that are not directly observable but reflect the essential characteristics of metabolism—are the core basis for generating metabolic profiles. Examples include the energy metabolism stability coefficient, which reflects the degree of fluctuation in energy consumption over time, the trend of cell hydration, and water use efficiency derived from water intake and skin temperature.

[0034] In some embodiments, metabolic profiling is a visual description of a user's metabolic state generated based on hidden variables, focusing on water metabolism, intake-expenditure balance, cellular hydration, and energy metabolism, with two main dimensions: basal metabolic rate and energy deficit / surplus. Example: The report shows a water balance index of +200ml, indicating intake > expenditure and a 20% decrease in cellular hydration; an energy deficit of -300kcal, indicating intake exceeds expenditure and a predicted basal metabolic rate of 1300kcal / day.

[0035] In some embodiments, metabolic profiles can be categorized based on the rate of nutrient absorption, specifically including but not limited to at least one of the following:

[0036] Metabolically sensitive: People who respond extremely quickly to interventions such as diet, exercise, and hydration. Their metabolic indicators, energy consumption rate, and fat breakdown efficiency can fluctuate significantly within hours to one day as their lifestyle changes change. They can easily enter a new metabolic homeostasis, but they are also prone to plateaus due to adaptation to a single intervention.

[0037] Judgment Criterion: The metabolic response index in the hidden variables output by the AI ​​model is greater than a certain threshold, for example, 0.7. The higher the index, the more sensitive the response. Feature Examples: Dietary Adjustment: Reducing carbohydrate intake by 200kcal at dinner results in a basal metabolic rate (BMR) fluctuation greater than a fluctuation threshold, for example, 5%, the next day; Post-Exercise Metabolism: After a 30-minute jog in the morning, the activity of lipase increases by 40% within 1 hour, and fat continues to break down during the post-exercise oxygen debt period, consuming an additional 100-150kcal.

[0038] Metabolic Response Delay Type: This group responds slowly to lifestyle interventions, requiring at least 3 days of high-intensity intervention, such as continuous calorie restriction and exercise, to trigger changes in metabolic indicators. Short-term adjustments are unlikely to disrupt their inherent metabolic homeostasis, but once broken, the homeostasis lasts longer. Key Indicator: The metabolic response index in the hidden variables output by the AI ​​model is <0.4; the lower the index, the slower the response. Example of Characteristics:

[0039] Dietary adjustments: Reducing carbohydrate intake by 200kcal at dinner for two consecutive days only increased BMR by 2%, such as from 1500 to 1530kcal / day; Post-exercise metabolism: After a 30-minute jog in the morning, the activity of fat-decomposing enzymes only increased by 15% after 4 hours. During exercise, the body relies on glycogen for energy, with fat accounting for only 20% of the energy supply. There was no significant and sustained fat burning after exercise.

[0040] Metabolic profiles based on the rate of fat burning can include:

[0041] Post-exercise fat consumption is categorized and associated with metabolic types: rapid fat consumption type, which is more compatible with metabolically sensitive type: moderate to low intensity exercise, heart rate 120-140 beats / min, 30 minutes can activate fat metabolism pathways, fat accounts for 50% of energy supply during exercise, compared to about 30% in the general population, and fat burning continues during the post-exercise oxygen debt period, burning an additional 100-150kcal.

[0042] Slow-burning fat metabolism type is often matched with metabolic boost type: low-to-moderate intensity exercise is difficult to activate fat metabolism, requiring high-intensity interval training, HIIT or more than 60 minutes of aerobic exercise. During exercise, it relies on glycogen for energy, with fat accounting for less than 20% of energy. There is no continuous fat burning after exercise, and it is necessary to rely on the cumulative intensity of exercise to force the breakdown of fat.

[0043] In some embodiments, fat gain / loss information is obtained directly from devices such as body fat scales and DXA scanners, or indirectly derived by AI models based on weight, diet, and exercise data, such as weekly body fat percentage changes and fat trend predictions. Example: A body fat scale shows that a user's body fat percentage increased by 0.3% this week, or an AI model predicts that if the current diet is maintained, fat mass may increase by 2 kg next month.

[0044] In some embodiments, the intervention strategy combines metabolic profiles with information on fat gain and loss to create a dynamic and personalized behavior adjustment plan for users, covering dimensions such as diet, exercise, and hydration, and supports real-time optimization.

[0045] In the S1110, wearable devices, smart hardware, and apps continuously collect multi-dimensional time-series data from users across all scenarios, covering core metabolic indicators. Example: Wearable devices, such as Apple Watch, collect heart rate (resting heart rate, exercise heart rate), daily steps, sleep stages (deep sleep duration, light sleep duration), and skin temperature range.

[0046] Body fat scale: Collects weight, body fat percentage, and muscle mass.

[0047] Diet app: Users enter their three meals a day. Breakfast: Oatmeal porridge 300kcal, protein 10g; Lunch: Steak 200kcal, fat 15g, water intake 1500ml per day.

[0048] Data integration: Generate time-series datasets by timestamp, such as heart rate 60 beats / min at 8:00 on July 1st; dietary calories 500kcal at 12:00 on July 1st.

[0049] In S1120, a time-series model is used to learn the temporal relationships and patterns of data, mapping the raw data into hidden variables that reflect the metabolic essence. Example: Input: The above multi-source time-series data, sorted by time, including heart rate, steps, diet, body fat percentage, etc.; AI model: LSTM network, which learns the relationships between heart rate fluctuations, exercise intensity, energy consumption, water intake, skin temperature, and cell hydration through training;

[0050] Output hidden variables: For example, metabolic fluctuation index, the higher the value, the more unstable the energy metabolism, indicating that the user's current energy consumption fluctuates greatly; water metabolism efficiency, the lower the value, the worse the water utilization, indicating that the cells do not absorb enough water after drinking.

[0051] In S1130, a metabolic profile is generated based on hidden variables. For example, abstract hidden variables are transformed into an intuitive and interpretable metabolic status report, focusing on the dual dimensions of water and energy metabolism. Example: A metabolic profile may involve one of the following parameters:

[0052] Water balance index: Adding 300ml, the intake of 2000ml vs. the expenditure of 1700ml indicates that the intake is slightly excessive and water utilization needs to be monitored.

[0053] Cellular hydration trend: decreased by 15%. Combined with skin conductivity and heart rate variability data, this suggests that water was not effectively absorbed by cells, possibly due to concentrated water intake.

[0054] Basal metabolic rate (BMR): Predicted value 1400 kcal / day, calculated based on metabolic rate trends in age, sex, muscle mass, and hidden variables; normal range 1200-1600 kcal.

[0055] Energy deficit: -200kcal. Intake of 1600kcal vs. expenditure of 1400kcal indicates an energy surplus, which may lead to fat accumulation in the long term.

[0056] Visual presentation: Generate a metabolic profile report with charts and text. If your energy metabolism is at risk of surplus or your water use efficiency is low, it is recommended to adjust your diet and water intake strategies.

[0057] In other embodiments, S1130 may include: an AI model distinguishing users as sensitive / sensitive types by metabolic response index, and combining exercise data, intensity, duration, and heart rate zone to deduce the classification of fat consumption rate after exercise, generating a more accurate metabolic characteristic report.

[0058] Example 1: Metabolically sensitive type plus rapid fat burning type

[0059] Hidden variable input: Metabolic response index = 0.85, calculated by weighting sub-variables such as BMR fluctuation after dietary adjustment, heart rate recovery speed after exercise, and cell metabolic pathway activation delay time; Exercise data: 30 minutes of jogging in the morning, heart rate 130 beats / min, low to moderate intensity.

[0060] In some embodiments, a metabolic profile includes at least three tags: a metabolic subtype tag, a post-exercise fat loss tag, and a core metabolic problem tag.

[0061] Example, metabolic profile output:

[0062] Metabolic typing label: Metabolic sensitive, response index 0.85. Characteristics of metabolic sensitive type: extremely rapid response to intervention, easy to quickly adjust metabolic state, but requires dynamic changes to avoid adaptation.

[0063] Post-exercise fat burning label: Fast fat burning type. Characteristics of fast fat burning type: The proportion of fat energy supply during exercise exceeds the first preset value, for example, 50%. For example, an additional 120kcal is burned within 2 hours after exercise, which is equivalent to continuous fat burning for 30 minutes after jogging.

[0064] Core metabolic issues can be tagged with water metabolism and energy metabolism.

[0065] Water metabolism label: Cellular hydration fluctuates greatly from day to day, for example, it is low in the morning and high in the afternoon, requiring dynamic adjustment of water replenishment;

[0066] Energy metabolism label: Basal Metabolic Rate (BMR) is sensitive to dietary changes. If you lose 200kcal yesterday, your BMR will increase by 8% today, requiring frequent fine-tuning of the calorie deficit.

[0067] Example 2: Metabolic euphoria combined with slow fat burning:

[0068] Hidden variable input: metabolic response index, for example, 0.32; exercise data: 30-minute jog in the morning, heart rate 125 beats / min, low to moderate intensity.

[0069] Metabolic profile output:

[0070] Metabolic typing label: Metabolic shock type, response index 0.32, indicating that it is a metabolic shock type and requires continuous high-intensity intervention to break the homeostasis. Short-term adjustments are unlikely to be effective and cumulative stimulation is required to trigger metabolic changes.

[0071] Post-exercise fat burning label: Slow-burning fat burning type.

[0072] In exercises characterized by slow fat burning, fat is used as the primary energy source, relying mainly on glycogen. There is no sustained fat burning after exercise, requiring increased intensity / duration to activate fat metabolism. For example, the second preset value is less than the first preset value. For instance, the second preset value could be 20%.

[0073] Core metabolic issues tags:

[0074] Water metabolism label: Strong cellular hydration homeostasis with daily fluctuations of <5%; a fixed hydration regimen needs to be followed for 2 weeks to see results.

[0075] Energy metabolism label: BMR is not sensitive to diet. A 2% increase in BMR occurs after 200kcal reduction for two consecutive days. Continued calorie control and exercise are needed to break through the plateau.

[0076] In S1140: Intervention strategies are generated by combining metabolic profiles with information on fat gain / loss. The correlation between metabolic status and fat changes is analyzed to customize personalized, dynamically adjustable behavioral plans covering dimensions such as diet, exercise, and hydration. Example: Fat gain / loss information: The body fat scale shows a 0.5% increase in body fat percentage this week; combined with an energy surplus, this confirms a trend of fat accumulation.

[0077] The derivation of intervention strategies may include, but is not limited to, at least one of the following:

[0078] Diet: Reduce refined carbohydrates. For example, replace rice with quinoa for dinner to reduce calories from 200kcal to 150kcal. Increase dietary fiber and protein intake. Add an egg to breakfast to increase protein by 7g and maintain muscle mass.

[0079] Exercise: 30 minutes of moderate-intensity aerobic exercise daily, such as brisk walking, with a target heart rate of 120-140 beats / minute, to increase energy expenditure and close the energy deficit; plus 10 minutes of strength training, such as squats, to maintain muscle mass and prevent a drop in BMR.

[0080] Hydration: Adjust your water intake in segments: drink 200ml 1 hour before exercise, 50ml every 15 minutes during exercise, and 100ml 30 minutes after meals to improve cell hydration;

[0081] Dynamic optimization: If the body fat percentage decreases by less than 0.2% next week, add HIIT training, replacing brisk walking with 15 minutes to improve fat burning efficiency; if water metabolism efficiency improves, fix the drinking times at 7:00 / 11:00 / 15:00 / 19:00.

[0082] S1140 may include: combining metabolic profiles with information on fat gain / loss to propose an intervention strategy adapted to the metabolic subtype. For example, the logic of this intervention strategy is as follows: targeting the rapid response characteristic of metabolically sensitive subtypes, dynamic fine-tuning is employed to avoid adaptation. Utilizing the characteristics of interventions being easily effective yet easily adapted to, lifestyle variables are frequently switched to continuously stimulate the metabolic system.

[0083] Dynamic adaptation to metabolic homeostasis: Sensitive individuals use dynamic changes to prevent metabolic adaptation by changing their diet and exercise patterns weekly; while those with sudden onset of symptoms use continuous stimulation to disrupt homeostasis by controlling calorie intake for four consecutive days and gradually increasing exercise intensity. These two logics cover the two extreme mechanisms of human metabolic regulation, truly achieving personalized solutions for each individual.

[0084] By incorporating metabolic typing and differences in post-exercise fat consumption, this method upgrades the focus from metabolic outcomes to understanding metabolic characteristics, making AI intervention more aligned with human biological principles—the sensitive type is like a finely tuned engine that requires dynamic maintenance; the acute type is like a heavy gear that needs powerful drive. This typing-based thinking is the core breakthrough in precision health management.

[0085] For delayed responses in metabolic shock-type disorders, continuous stimulation is used to disrupt metabolic homeostasis. For example, leveraging the cumulative effect of interventions, a fixed high-intensity regimen is implemented for 3-5 days, forcibly disrupting metabolic homeostasis through continuous stimulation. Example, in a fat loss scenario:

[0086] Dietary intervention: Fixing a low-GI template:

[0087] First, for four consecutive days, breakfast consisted of eggs and whole-wheat bread; lunch consisted of brown rice and chicken breast; and dinner consisted of broccoli and shrimp, with strict calorie control, reducing calories by 500kcal per day. Then, on the fifth day, a cheat meal was introduced, adding 300kcal to the diet, with options such as pizza or hamburgers, to trigger a metabolic reboot.

[0088] Metabolic stagnation requires continuous and stable intervention to accumulate an energy deficit. For example, controlling calorie intake for four consecutive days until glycogen reserves are depleted can then trigger fat breakdown. Cheat meals can prevent long-term low metabolic adaptation.

[0089] Progressive high-intensity exercise should be used during exercise intervention:

[0090] Week 1: 30 minutes of jogging 3 times a week, aiming for a heart rate of 120-130 bpm. Week 2: 20 minutes of high-intensity interval training (HIIT) 3 times a week, aiming for a heart rate of 150-170 bpm, plus 10 minutes of strength training, including squats / planks. Week 3: 25 minutes of HIIT 4 times a week, gradually increasing the intensity to forcibly activate fat metabolism pathways.

[0091] For low- to moderate-intensity exercise that doesn't trigger fat metabolism (such as jogging for 30 minutes where fat may only provide 20% of the energy), HIIT can force the breakdown of fat through oxygen debt after anaerobic exercise.

[0092] Hydration intervention: A fixed time and quantity plan: Drink 300ml of water at 7:00 / 11:00 / 15:00 / 19:00, containing trace electrolytes, for two consecutive weeks. Once the metabolic system adapts to the new hydration pattern, adjust to drinking 200ml 20 minutes before exercise and 50ml every 15 minutes during exercise to optimize cellular hydration. This approach fosters a strong homeostasis of metabolically active cellular hydration. If fixed hydration for three days results in no change in hydration, two weeks of continuous hydration are needed to break the old homeostasis and further optimize details to improve efficiency.

[0093] In some embodiments, intervention actions are performed and dynamic feedback is provided. For example, wearable devices or app push notifications guide users to execute strategies, record behavioral data in real time, and feed it back to the AI ​​model to achieve dynamic optimization of the strategy.

[0094] Example: Smart bracelet / APP reminder: 7:00: Today's water intake target is 2000ml, it is recommended to drink 300ml at 7:30 to improve cell hydration in the morning;

[0095] 12:00: Lunch recommendation: Quinoa rice with steamed fish, 400kcal, 25g protein;

[0096] 15:00: 30 minutes of brisk walking training, current heart rate should be maintained at 120-140 beats / minute, real-time display shows current heart rate is 135 beats / minute, target achieved;

[0097] Data feedback and optimization: If a user does not complete the exercise for two consecutive days, the AI ​​will automatically adjust the strategy and switch to 15 minutes of HIIT to adapt to time-constrained scenarios;

[0098] If the water intake is sufficient but the cell hydration level is still low, the model analyzes the water temperature and suggests variables such as warm water if ice water is frequently consumed, and optimizes the hydration plan.

[0099] In some embodiments, by utilizing multi-source heterogeneous data and hidden variables, subtle metabolic changes can be accurately captured, such as a weight loss plateau caused by decreased water metabolism efficiency. For example, if a user achieves their energy deficit but their body fat percentage does not decrease, AI can identify the correlation between poor water utilization and a decreased metabolic rate, and adjust hydration strategies to overcome the bottleneck.

[0100] Personalized and deeply customized: Metabolic profiles and intervention strategies are generated entirely based on individual data and are tailored to lifestyle habits. For example, if a user dislikes running, AI can replace it with swimming, cycling, or other sports; if a user enjoys carbohydrates, the strategy can be adjusted to include low-GI carbohydrates at optimal times to improve adherence.

[0101] Dynamic adaptive optimization: With continuous input of time-series data, the AI ​​model learns and provides feedback in real time. For example, if dietary disorders occur during travel, the short-term strategy is automatically adjusted to ensure that the intervention effect does not diminish with changes in the life scenario.

[0102] Comprehensive metabolic coverage: Simultaneously manages both water and energy metabolism, overcoming the shortcomings of traditional methods that only control calories while neglecting water. For example, by improving cellular hydration and promoting lipase activity, energy consumption efficiency is enhanced. Experiments have shown that after optimizing water metabolism, users' basal metabolic rate increases by an average of 5-8%.

[0103] Through the above embodiments, this AI intervention method realizes a closed loop from data collection to precise intervention, providing a feasible technical path for personalized health management, such as weight loss and chronic disease prevention.

[0104] In some embodiments, S1110 may include at least one of the following:

[0105] The weighing device acquires the target user's weight and / or body fat information.

[0106] The device acquires at least one of the target user's heart rate information, activity level information, and sleep information collected by the limb-worn device.

[0107] Obtain the metabolic information of the target user collected by the blood glucose meter;

[0108] Use head-mounted devices to collect appetite information and / or emotional state;

[0109] The target user's dietary information is obtained through diet-related applications.

[0110] Weighing equipment: Smart hardware with functions such as weight, body fat percentage, muscle mass, and water percentage detection, such as body fat scales and medical weighing scales. It collects body composition data through technologies such as bioelectrical impedance and pressure sensing, and outputs time-series body composition information, such as the user's weight value plus body fat percentage at 7:00 a.m. every day.

[0111] Wearable devices for limbs: Smart devices worn on the wrists, ankles, and other limbs, such as smart bracelets, sports watches, and electromyography sensors. They collect time-series data such as heart rate fluctuations, exercise volume, steps / exercise intensity / duration, sleep stages, and deep sleep / light sleep / REM sleep stages through photoelectric heart rate sensors, accelerometers, and sleep monitoring modules.

[0112] Blood glucose meter: A device that collects blood glucose concentration data through fingertip / dynamic monitoring by blood sampling. It includes traditional blood glucose meters and dynamic glucose monitoring systems (CGM). It outputs key indicators that reflect the state of energy metabolism, such as fasting blood glucose, postprandial blood glucose fluctuation, and area under the blood glucose curve, which are used to infer hidden variables such as carbohydrate metabolism efficiency and insulin sensitivity.

[0113] Head-mounted devices for appetite / emotion data acquisition: Head-mounted devices integrating EEG, heart rate variability, HRV modules, and facial expression recognition, such as smart headbands and AR glasses, analyze brainwaves, such as hypothalamic appetite-related signals and autonomic nervous activity, HRV reflecting emotional stress and facial micro-expressions, to collect unstructured data such as appetite impulse intensity, emotional state, anxiety / pleasure, etc., and convert them into quantifiable temporal features, such as a user's appetite index at 12:00 = 80.

[0114] Food-related applications: Apps such as Mint Health, which allow users to actively input their diet information, including ingredients, calories, and meal times, or automatically analyze diets through image recognition, and output structured dietary data, such as breakfast on July 1st at 8:00 AM: oatmeal porridge, 300kcal, 50g carbohydrates, to supplement the source of energy intake.

[0115] In some embodiments, step S1110: Multimodal device collaboration to construct a metabolic data closed loop. Through the combination of body composition detection, weighing equipment, physiological signal monitoring, limb wearable and headband devices, metabolic index quantification, blood glucose meter, dietary behavior recording, and APP, the system covers four dimensions: energy input, diet, energy consumption, exercise / basal metabolism, metabolic intermediates, blood glucose, behavioral drivers, and appetite / mood, providing complete data support for subsequent metabolic analysis.

[0116] For example, the body fat scale uses bioelectrical impedance pressure sensing technology to detect body fat and weight.

[0117] The body fat scale is designed to collect data from the user standing barefoot on an empty stomach in the morning. The scale automatically synchronizes the data to the cloud and outputs time-series body composition data.

[0118] Data value: Weight fluctuations reflect short-term water / energy balance;

[0119] Body fat percentage plus muscle mass is used to dynamically calculate basal metabolic rate (BMR), such as the simplified formula: BMR = muscle mass × 15 plus body weight × 2.

[0120] Visceral fat levels indicate a risk of metabolic syndrome; levels >7 require intensive intervention.

[0121] In some embodiments, wearable devices for limbs may include smart bracelets, smartwatches, or smartwatches.

[0122] For example, wearable devices for the limbs may include, but are not limited to, photoelectric heart rate sensors, six-axis sensors, and sleep monitoring.

[0123] Wearable devices for limbs collect heart rate information: real-time monitoring, 1 minute / time, outputting heart rate fluctuation curves during exercise.

[0124] Exercise data: Energy consumption is calculated based on steps and exercise intensity, and exercise time-series data is output.

[0125] Sleep information: Nighttime sleep stage monitoring, outputting sleep cycle data:

[0126] Data value: Heart rate variability reflects autonomic nervous system balance and stress / recovery status; exercise volume accurately predicts energy deficit, intake-expenditure; sleep stages are associated with growth hormone secretion, and insufficient deep sleep can lead to decreased fat breakdown efficiency and inhibited muscle synthesis.

[0127] For example, a dynamic glucose monitoring system using a blood glucose meter can be used. For instance, the blood glucose meter may include a subcutaneous sensor and can perform periodic monitoring, such as monitoring every 5 minutes, to continuously monitor blood glucose fluctuations and output a blood glucose time-series curve plus key metabolic indicators.

[0128] The amplitude of blood glucose fluctuations reflects insulin sensitivity; large fluctuations indicate a high risk of insulin resistance. Postprandial blood glucose peaks are related to carbohydrate metabolism efficiency; high peaks require adjustment of the dietary glycemic index (GI value), such as by replacing it with low-GI staple foods. Blood glucose provides real-time fuel status for energy metabolism models; blood glucose = the current fuel tank capacity for energy metabolism.

[0129] Head-mounted devices may include, but are not limited to, smart glasses, smart headphones, or smart headbands.

[0130] Appetite Information: Before meals, at 11:30 AM, brainwaves in the hypothalamus region were monitored using a head-mounted device. The intensity of appetite-related delta waves was analyzed and combined with historical eating habits to output an appetite index.

[0131] Emotional state: Through HRV spectrum analysis and LF / HF ratio, reflecting the balance of sympathetic / parasympathetic nerves, combined with facial micro-expression recognition and AR glasses assistance, emotional labels are output.

[0132] Appetite index can warn of the risk of emotional eating. For example, a spike in appetite during anxiety requires early intervention in diet; emotional state is associated with stress-induced metabolic disorders. For instance, chronic anxiety may lead to elevated cortisol levels and visceral fat accumulation, filling a data gap in behavioral drivers.

[0133] The head-mounted device measures the target user’s appetite and / or mood before and after mealtimes.

[0134] In some embodiments, users manually enter their dietary information, or image recognition assists in obtaining this information. For example, food delivery apps or shopping apps can be used to obtain users' dietary information.

[0135] Dietary calories plus the ratio of macronutrients are used to calculate the energy deficit, intake minus expenditure;

[0136] The glycemic index (GI) is associated with blood sugar fluctuations. For example, a low GI indicates stable blood sugar levels and leads to sustained fat breakdown.

[0137] Precise components that provide energy input to metabolic models, such as a high-protein diet, can boost basal metabolism.

[0138] In this embodiment, five types of devices / applications cover energy input, consumption, metabolic intermediates, body composition, and behavioral-driven closed loops. For example, a user's large blood sugar fluctuations, high appetite, and emotional anxiety can lead to a vicious cycle of stress-induced binge eating, blood sugar disorders, and decreased metabolic rate, allowing for precise identification of intervention points. For instance, prioritizing anxiety relief rather than simply controlling calorie intake.

[0139] In some embodiments, multimodal data provides richer features for AI models. For example, by combining sleep stages, deep sleep duration, blood glucose fluctuations, and muscle mass, the basal metabolic rate (BMR) can be dynamically calculated. Traditional formulas assume a stable BMR, but in this embodiment, due to changes in sleep / muscle mass, the weekly BMR can fluctuate by ±10%. By combining appetite index, dietary GI value, and peak blood glucose levels, a carbohydrate metabolism sensitivity coefficient can be extracted, reflecting the body's efficiency in breaking down carbohydrates.

[0140] In some embodiments, when a user's body fat percentage increases, traditional methods only indicate a calorie surplus. This disclosure allows for data linking to trace the source: if a diet app accurately records calories, but exercise monitoring devices show insufficient calorie expenditure, then exercise requirements or reminders may be added to the intervention strategy. Similarly, if a blood glucose meter shows large fluctuations in blood sugar, insulin resistance, and accelerated fat synthesis, then dietary adjustments may be made through intervention strategies. Furthermore, if a head-mounted device monitors emotional anxiety, elevated cortisol, and visceral fat accumulation, then emotional intervention can be introduced, upgrading from outcome intervention to root cause resolution.

[0141] In this embodiment of the disclosure, in addition to traditional dietary intervention and exercise intervention, the intervention strategy also incorporates emotional intervention, thereby achieving comprehensive weight control.

[0142] In some embodiments, the data characteristics of different metabolic subtypes, such as sensitive and sensitized types, differ significantly. Multimodal data allows AI models to more accurately identify subtypes and adapt to differentiated intervention strategies. For example, sensitive types require dynamic dietary adjustments, while sensitized types require continuous high-intensity stimulation.

[0143] Through multi-dimensional data collection in S1110, this embodiment of the disclosure constructs a complete data ecosystem of input, consumption, metabolism, and behavior, providing sufficient raw material for subsequent extraction of hidden variables and generation of metabolic profiles. The more complete the data, the deeper the AI ​​insights, and the closer the intervention is to the individual's physiological essence, truly realizing the leap from general health management to precision individual medicine.

[0144] In some embodiments, S1120 may include:

[0145] The multi-source heterogeneous data is subjected to data standardization processing to obtain a three-dimensional data matrix; wherein, the three-dimensional data matrix includes a first matrix and a second matrix; the first matrix includes time-series data extracted from the multi-source heterogeneous data; the second matrix includes non-time-series data extracted from the multi-source heterogeneous data;

[0146] Temporal features are extracted from the first matrix based on a time correlation algorithm; the temporal features include at least one of local temporal features and global temporal features;

[0147] Extract non-temporal features from the second matrix based on hot coding or embedding layer transformation;

[0148] Based on the synergistic effect of dynamic physiological processes and static physiological states, the hidden variables are obtained by fusing the temporal features and the non-temporal features; the temporal features are used to reflect the dynamic physiological processes of the target user; the non-temporal features are used to reflect the static physiological states of the target user.

[0149] In some embodiments, the first matrix is ​​generated based on at least one of the following information: heart rate information, exercise volume information, sleep information, and metabolic information; and / or, the second matrix is ​​generated based on at least one of the following information: weight information, body fat information, appetite information, emotional state, and dietary information.

[0150] In some embodiments, any element in the three-dimensional data matrix may include numerical values ​​in three dimensions. The standardized data structure splits into a time-series matrix, a first matrix, and a non-time-series matrix, a second matrix, achieving separation of dynamic and static data. The time-series matrix stores continuous data that changes over time, such as hourly heart rate and daily weight. The non-time-series matrix stores static / discrete data, such as gender, genetic history, and device model. To achieve the data attribute characteristics of a three-dimensional matrix in the second matrix, the non-time-series matrix is ​​automatically expanded into a three-dimensional matrix before being input into the AI ​​model; for example, by introducing 0 corresponding to the time dimension, thus reducing misunderstandings caused by introducing other numerical values ​​in the time dimension.

[0151] In some embodiments, time correlation algorithms for mining time dependencies in time series data are divided into two categories:

[0152] Local temporal features: Short-term patterns, such as 5-minute heart rate variability, are extracted using a sliding window.

[0153] Global temporal features: Long Short-Term Memory (LSTM) and / or Transformer are used to capture long-term trends, such as the 24-hour heart rate-diet association.

[0154] In some embodiments, hot coding (e.g., One-Hot) converts low-cardinality categorical data, such as gender and genetic history, into binary vectors to avoid numerical ambiguity.

[0155] In some embodiments, the embedding layer maps high cardinality classification data, such as diet type and exercise type, into low-dimensional dense vectors while preserving the semantic association between categories. For example, brown rice and oats are close in vector distance because they are both low-GI staple foods.

[0156] In some embodiments, hidden variables are potential parameters that reflect the essence of metabolism after integrating dynamic, temporal and static, and non-temporal features. These parameters cannot be directly observed, but they determine the core dimensions of the metabolic profile, such as energy metabolism stability and cell hydration trends.

[0157] In some embodiments, time alignment and interpolation are performed by interpolating weight and daily updates to hourly data and aligning heart rate to minute-level timestamps, and then averaging the data hourly.

[0158] In some embodiments, the classification data is encoded as follows: diet type, brown rice: converted into a One-Hot vector using heat encoding, assuming only low-GI staple food / high-GI staple food / protein / fat are considered, brown rice is encoded as [1, 0, 0, 0]; gender, male: heat encoding is [1, 0]; genetic history, no diabetes is encoded as [1, 0], and diabetes is encoded as [0, 1].

[0159] Using algorithms to mine dynamic physiological patterns in time-series data, including local fluctuations and global trends. Local time-series feature extraction: sliding window with statistics: window setting: 5-minute window, capturing micro-fluctuations in heart rate before / after meals.

[0160] Feature calculation: Taking the period around 12:00 lunch, from 11:55 to 12:05 as an example, the heart rate sequence is [85, 90, 88]. After normalization, the following calculation is performed:

[0161] Volatility = Standard Deviation / Mean. The lower the volatility, the more stable the heart rate.

[0162] Peak offset = maximum value - mean / mean. Peak offset reflects the intensity of dietary stimulation on heart rate.

[0163] Global temporal feature extraction utilizes a transformation model with a self-attention mechanism. The self-attention mechanism of the Transformer model identifies the midday heart rate peak, a strong correlation between 12:00 and dietary coding [1, 0, 0, 0], the nighttime heart rate trough, and the association between 2:00 and the deep sleep stage.

[0164] Output: A 128-dimensional global trend vector containing time-dependent patterns of heart rate, diet, and sleep, such as the associated weights for low-GI diet, small heart rate fluctuations, and stable energy metabolism.

[0165] Static data is transformed into numerical features that can participate in model calculations while preserving category semantics. Low-cardinality classification is handled using hot encoding: gender (male); genetic history (no diabetes); device model (wristband A), assuming three devices: wristband A / B, body fat scale C.

[0166] Embedding layers handle high cardinality classification, including food categories:

[0167] In the AI ​​model embedding layer, it will automatically identify that the vector distance between brown rice and oats is less than the distance between brown rice and steak, because the former two are both low-GI staple foods and are semantically closer.

[0168] Non-temporal feature fusion: numerical features, height 1.75, hot-coded features, gender [1,0], embedded features, and diet are spliced ​​together to form a static feature vector.

[0169] Dynamic and static features are integrated to generate hidden variables. For example, based on the synergistic effect of dynamic physiological processes and time-series features with static physiological states and non-time-series features, neural networks are used to fuse features and output hidden variables that reflect the essence of metabolism.

[0170] Before standardization, the data format was chaotic and the frequency conflicted, such as daily weight updates versus minute-level heart rate updates. After standardization, a three-dimensional matrix was formed, allowing dynamic data, heart rate fluctuations and static data, gender / height to be in their proper places—dynamically capturing real-time changes in metabolism, and statically solidifying the basic metabolic threshold.

[0171] In this embodiment of the disclosure, the spatiotemporal penetration of feature extraction is as follows: Local temporal features: capturing instantaneous metabolic fluctuations, such as a slight increase in heart rate before meals to warn of the risk of overeating; Global temporal features: capturing the metabolic patterns throughout the day, such as the correlation between the trough of heart rate during deep sleep and fat breakdown efficiency; Non-temporal features: using an embedding layer to allow food categories to carry nutritional semantics, such as automatically classifying low-GI staple foods into one category, avoiding the dimensionality curse of traditional hot coding.

[0172] Hidden Variables in Metabolic Essence: Traditional methods describe metabolism using only weight changes. This disclosure combines dynamic and static methods to accurately pinpoint the low cellular hydration caused by insufficient water intake, which in turn leads to a decrease in metabolic rate, allowing for more direct intervention.

[0173] Hidden variables are the raw materials for metabolic profiling—energy metabolism stability corresponds to the energy module, cell hydration corresponds to the water module, and insulin sensitivity corresponds to the glucose metabolism module, allowing subsequent profiling to upgrade from describing phenomena to explaining the essence. For example, if energy metabolism is stable but water is inefficient, then the drinking strategy needs to be adjusted rather than controlling calories.

[0174] In some embodiments, the extraction of temporal features from the first matrix based on the time correlation algorithm includes:

[0175] A sliding first time window is used to extract a first water metabolism feature matrix from a portion of the first matrix corresponding to the heart rate information; wherein, the first time window is related to the human body's water metabolism cycle; and the first water metabolism matrix is ​​related to the heart rate information.

[0176] Process a portion of the first matrix corresponding to the sleep information to obtain a sleep state transition matrix and a second water metabolism feature matrix; the second water metabolism matrix is ​​related to water metabolism during the sleep cycle.

[0177] Multi-scale feature decomposition is used to process a portion of the first matrix related to the metabolic information, and a temporal anchoring mechanism is employed to obtain the first energy metabolism feature matrix; the temporal anchoring mechanism is triggered by the dietary event corresponding to the dietary information.

[0178] Dynamic Time Warping (DTW) is used to extract the temporal feature matrix relating exercise status and fat breakdown from a portion of the first matrix related to the exercise information.

[0179] A third water metabolism feature matrix is ​​obtained by processing a portion of the first matrix related to the motion information using a second time window; the third water metabolism matrix is ​​related to the target user's exercise intensity or instantaneous energy consumption.

[0180] The first matrix related to the motion information is processed using a third time window to obtain the energy metabolism feature matrix.

[0181] These features can be further used to extract hidden variables to ensure that the hidden variables accurately reflect the target user's state.

[0182] In some embodiments, the extraction of non-temporal features from the second matrix based on hot coding or embedding layer transformation includes:

[0183] The weight range is converted into a one-hot feature matrix through hot encoding, or mapped to a low-dimensional feature matrix through an embedding layer, reflecting the weight baseline state feature matrix; the weight baseline state feature matrix is ​​related to the weight information.

[0184] A hot coding method is used to distinguish the classification labels of visceral fat and subcutaneous fat. These labels are then converted into feature vectors using an embedding layer to obtain metabolic difference feature matrices for different fat types. The metabolic difference feature matrices are related to the constitution information.

[0185] Different appetite levels are converted into unique hot codes or quantized feature matrices via an embedding layer through hot coding.

[0186] Different emotion types are represented using hot encoding, and then converted into an emotion-related metabolic feature matrix through an embedding layer; the emotion-related metabolic feature matrix is ​​used to reflect the potential impact of emotions on metabolism and is related to the emotional state;

[0187] Dietary structure types are labeled by hot encoding and combined with an embedding layer to decompose them into a holistic pattern feature matrix of energy intake that maps nutritional balance features; the holistic pattern feature matrix is ​​related to the mapping information.

[0188] In summary, by employing a feature transformation strategy involving hot encoding and embedding layers, we have achieved a precise and structured representation of discrete health information. We have uncovered deep correlations among multi-dimensional features such as weight, fat, mood, and diet, providing fine-grained data support for personalized health management (metabolic analysis, emotional intervention, dietary optimization, etc.). At the same time, we have improved the model's processing efficiency through low-dimensional feature mapping, ultimately driving the evolution of health data analysis from extensive correlation to precision, personalization, and efficiency.

[0189] In this embodiment of the disclosure, the weight baseline includes three dimensions: body mass index, muscle mass, and body fat distribution.

[0190] The Body Mass Index (BMI) is calculated as weight / height, reflecting the ratio of total weight to height. It is used for preliminary screening of abnormal weight (underweight / overweight / obese).

[0191] Muscle mass: refers to the total weight of muscle tissue in the body (including skeletal muscle, smooth muscle, etc.), and is usually measured by equipment such as bioelectrical impedance analysis (BIA) and dual-energy X-ray absorptiometry (DXA).

[0192] Introducing muscle mass into a weight baseline can correct misjudgments caused by BMI. For example, users with high muscle mass may have lower metabolic health risks than users with high body fat at the same BMI, even if their BMI is above the standard.

[0193] Body fat distribution: the proportion of fat deposited in different parts of the body. For example, the distribution of visceral fat, subcutaneous fat, and abdominal / hip / limb fat needs to be assessed using CT, MRI, or body fat scale algorithm models (such as fat distribution estimation based on electrical impedance).

[0194] Visceral fat (fat surrounding organs such as the liver and pancreas) has a far greater impact on metabolic health than subcutaneous fat—excessive visceral fat is directly linked to insulin resistance, fatty liver, and the risk of cardiovascular disease, while body fat distribution can precisely capture this risk.

[0195] A weight baseline composed of BMI, muscle mass, and body fat distribution breaks through the limitations of traditional single-indicator weight measurements. Traditional weight baselines only focus on the numerical value of weight, but:

[0196] Two people with the same weight may have vastly different health risks: one may have a healthy, muscular build, while the other may have a high-risk, fatty build. Even if their weight and BMI are the same, the health risks can be drastically different depending on whether fat is deposited in the abdomen (more visceral fat) or in the buttocks (more subcutaneous fat).

[0197] Users can be categorized using a three-dimensional baseline:

[0198] Type A: High BMI, high muscle mass, and healthy body fat distribution may indicate muscular overweight, such as athletes, with low metabolic risk.

[0199] Type B: Normal BMI, low muscle mass, and high visceral fat indicate a hidden metabolic risk. If the weight is normal but the fat distribution is abnormal, special intervention is needed.

[0200] Type C: High BMI, low muscle mass, and high visceral fat indicate a typical high risk of obesity, requiring comprehensive weight control, muscle gain, and fat control.

[0201] During the intervention process, the effectiveness of the strategy can be verified through changes in three dimensions:

[0202] If the goal is to reduce visceral fat, it is necessary to monitor the decrease in the proportion of visceral fat in body fat distribution (rather than just looking at weight / BMI).

[0203] If the goal is to gain muscle and weight, it is necessary to ensure that while muscle mass increases, body fat distribution does not worsen (avoid "gaining muscle but gaining fat").

[0204] The AI ​​model can use this three-dimensional baseline as input features to train the mapping relationship between abnormal weight baseline, metabolic risk and intervention strategies, thus avoiding one-size-fits-all interventions caused by a single BMI indicator.

[0205] In some embodiments, through feature engineering in S1120, multi-source data undergoes a transformation from raw materials and structured features to hidden variables, constructing a dynamic-static collaborative and local-global metabolic analysis for AI intervention methods. The more accurate the subsequent metabolic profile and intervention strategy, the more intelligent closed loop from data to health management can be achieved.

[0206] In some embodiments, the hidden variable obtained by fusing the temporal features and the non-temporal features based on the synergistic effect of dynamic physiological processes and static physiological states includes at least one of the following:

[0207] By fusing temporal features of heart rate and non-temporal features of body fat, a first hidden variable is obtained; the first hidden variable co-encodes the dynamic fluctuations of water metabolism and the basal metabolic differences of fat storage types.

[0208] By integrating temporal features of exercise volume and non-temporal features of diet, a second hidden variable is obtained; the second hidden variable is used to express the synergistic relationship between the target user's dynamic energy metabolism, the post-exercise fat consumption effect, and silver snake energy intake.

[0209] By integrating sleep temporal features and emotional non-temporal features, a third hidden variable is obtained; this third hidden variable is used to reflect the sleep-emotion co-metabolic effect.

[0210] By integrating metabolic time-series features and dietary non-time-series features, a fourth hidden variable is obtained; the fourth hidden variable is used to reflect dietary motivation, metabolic status, and appetite-driven energy input bias information.

[0211] By integrating temporal features of exercise and non-temporal features of body weight, a fifth hidden variable is obtained; this fifth hidden variable is used to reflect the personalized adaptation of exercise intervention to the body weight baseline.

[0212] This disclosure captures real-time metabolic fluctuations by leveraging the temporal characteristics of dynamic physiological processes, such as the temporal changes in heart rate and exercise intensity, and the non-temporal characteristics of static physiological states. It anchors basic metabolic attributes, such as body type and dietary baselines, through synergistic modeling, and uncovers the interaction mechanisms between these two types of features, such as how body constitution modulates the effect of heart rate fluctuations on water metabolism. This generates hidden variables focusing on different metabolic sub-dimensions. Each variable is defined by a threshold range, rather than specific numerical values, quantifying the health-disorder state of metabolic synergy and providing underlying logical support for precise intervention.

[0213] In some embodiments, a first hidden variable is generated based on temporal heart rate features and non-temporal body fat features.

[0214] On the dynamic side, heart rate temporal features: extract heart rate fluctuations, such as the magnitude of postprandial heart rate spikes, global trends, and diurnal heart rate fluctuation patterns, to characterize the dynamic processes of water metabolism and energy consumption;

[0215] On the static side, non-temporal characteristics of body fat: anchored to body fat percentage, muscle mass, body type, endomorph / ectomorph, etc., reflecting the static baseline of fat storage and basal metabolism;

[0216] Synergistic mechanism: The attention network is used to assign weights to key periods of heart rate fluctuations after meals / exercises, reflecting the body's regulation of heart rate-metabolism, such as the tendency of endomorphic body types to store fat due to heart rate fluctuations.

[0217] In some embodiments, the first hidden variable may be the water metabolism-fat storage synergy index, denoted as HV1. The first hidden variable can be used to quantify how body type modulates the effect of heart rate fluctuations on water metabolism and fat storage, reflecting the synergistic effect between dynamic heart rate and static body type.

[0218] In some embodiments, the baseline synergistic threshold is the lowest effective regulatory threshold for the dynamic-static synergistic effect of the metabolic system. Below this threshold, the body's ability to regulate heart rate and metabolism is extremely weak, which is common in healthy young people or in a state where metabolic synergy has not yet been initiated.

[0219] In some embodiments, the healthy synergistic range is: when HV1 is in this range, the synergistic effect between heart rate fluctuations and body type is benign. For example, in endomorphic individuals, the postprandial heart rate rises slightly but water metabolism remains stable, and fat storage is regulated by body type but not out of control.

[0220] In some embodiments, the metabolic disorder range: when HV1 is above this range, a synergistic effect triggers metabolic damage, such as a sudden increase in heart rate after a meal in endomorphic individuals, leading to a sharp drop in water metabolism efficiency and a significantly increased risk of fat accumulation.

[0221] This disclosure allows for the localization of interactive damage related to body constitution, heart rate, and metabolism. By defining threshold ranges, the healthy boundaries of metabolic synergy are clearly defined, enabling interventions to be precisely designed to target disordered regions, such as targeted postprandial hydration and low-GI diets to stabilize heart rate fluctuations.

[0222] In some embodiments, a second hidden variable is generated based on temporal features of exercise volume and non-temporal features of diet.

[0223] On the dynamic side, the temporal characteristics of exercise volume are extracted: exercise intensity, HIIT / daily steps, afterburn effect, and energy consumption residue 24 hours after exercise to characterize the dynamic energy consumption pattern.

[0224] On the static side, dietary non-temporal characteristics: anchored to the proportion of macronutrients, protein / carbohydrate / fat, dietary GI value, reflecting the structural baseline of energy intake;

[0225] Synergistic Mechanism: Using a gated recurrent unit (GRU), the GRU learns the temporal dependence of exercise-diet-energy metabolism. For example, a high-protein diet after HIIT can accelerate muscle synthesis and increase fat breakdown efficiency, capturing the synergistic effect between exercise intensity and dietary structure.

[0226] In some embodiments, the second hidden variable may be referred to as the dynamic energy metabolism synergy index, denoted as HV2.

[0227] In some embodiments, the synergistic effect of exercise intensity and dietary structure is quantified to reflect the fit between dynamic energy expenditure and static energy intake.

[0228] In some embodiments, the energy adaptation threshold is the minimum threshold for the exercise-diet synergy. For example, below the minimum threshold, there is no significant synergy between exercise intensity and dietary structure, and energy metabolism remains at a basal level.

[0229] In some embodiments, the efficient synergistic range is: when HV2 is in this range, exercise-diet synergistic effect is enhanced, for example, high-protein diet after HIIT, prolonged afterburn effect, and stable expansion of energy deficit.

[0230] In some embodiments, the inefficient synergy range is defined as follows: when HV2 is above this range, the synergy effect is unbalanced, for example, excessive exercise combined with a high-carbohydrate diet, which leads to a sharp drop in residual energy expenditure and metabolic fatigue, triggering the risk of binge eating.

[0231] By using threshold ranges, we can distinguish between synergistic and unbalanced states, enabling interventions to optimize the exercise-diet combination. For example, we can maintain high-efficiency HIIT with high protein intake and adjust exercise intensity or dietary structure in low-efficiency ranges.

[0232] In some embodiments, a third hidden variable is generated based on sleep temporal features and emotional non-temporal features.

[0233] On the dynamic side, sleep time sequence characteristics: extract sleep stages, deep sleep / light sleep / REM period, and awakening frequency to characterize the dynamic process of metabolic repair;

[0234] On the static side, non-temporal characteristics of emotion: anchored anxiety score and cortisol level, reflecting the static baseline of stress;

[0235] Synergistic mechanism: Using graph neural networks (GNNs), we model the associated pathways of sleep stages, emotional states, and metabolic hormones to capture the synergistic inhibitory effects of light sleep increasing anxiety, persistently elevated cortisol, and decreased metabolic rate.

[0236] The third hidden variable, also known as the sleep-emotion co-metabolic index, is denoted as HV3. This third hidden variable quantifies the co-metabolic effect between sleep quality and emotional state, reflecting the interaction between dynamic sleep repair and static emotional stress.

[0237] In some embodiments, the metabolic repair threshold is the minimum effective threshold for the synergistic repair of sleep and mood. For example, values ​​below the minimum effective threshold indicate that the synergistic repair effect of sleep and mood on metabolism is extremely weak, which is common in short-term sleep deprivation or mild anxiety.

[0238] In some embodiments, the homeostatic synergistic range is defined as follows: when HV3 is within this range, sleep and mood work together to maintain metabolic homeostasis. For example, sufficient deep sleep combined with low anxiety leads to normal growth hormone secretion and controllable cortisol fluctuations.

[0239] In some embodiments, the damage synergistic range: when HV3 is above this range, the synergistic effect triggers metabolic damage, for example, insufficient light sleep increases anxiety, leading to a sharp drop in growth hormone secretion, a sustained increase in cortisol and a decrease in basal metabolic rate.

[0240] By defining the threshold range, the boundary between metabolic homeostasis and damage can be clarified, enabling interventions to be upgraded into a synergistic approach to improve deep sleep and stress management, such as meditation to prolong deep sleep and mindfulness to reduce cortisol.

[0241] In some embodiments, a fourth hidden variable is generated based on metabolic time-series features and dietary non-time-series features.

[0242] On the dynamic side, metabolic time-series characteristics: extracting the amplitude of blood glucose fluctuations, the difference between fasting and postprandial peak values, peak delay, and the time to peak blood glucose after consuming sweets, thus characterizing the dynamic disorder of energy metabolism;

[0243] On the static side, non-temporal characteristics of eating: anchoring the appetite impulse index, the desire to eat during anxiety, the proportion of sweets, and the static baseline reflecting eating motivation;

[0244] Synergistic Mechanism: Using attention combined with LSTM to capture the closed loop of appetite urges, sweet food intake, blood sugar fluctuations, and more impulsive eating, revealing the metabolic driving logic of emotional eating. For example, anxiety triggers high sugar intake, and blood sugar disorders in turn fuel appetite urges.

[0245] The fourth hidden variable, also known as the eating motivation-metabolic bias co-existence index, is denoted as HV4. The fourth hidden variable represents the co-existence bias between quantitative eating motivations, such as appetite urges and metabolic states, such as blood glucose fluctuations, reflecting the metabolic drive loop of emotional eating.

[0246] In some embodiments, the metabolic balance threshold is the minimum fluctuation line of the diet-metabolism synergy deviation. Below this value, the synergy deviation between appetite urges and blood glucose fluctuations is minimal, and metabolism is basically stable.

[0247] Controllable deviation range: When HV4 is in this range, the synergistic deviation can be regulated by the metabolic system. For example, occasional anxiety eating may cause temporary fluctuations in blood sugar, and the appetite will return to normal as blood sugar recovers.

[0248] Out-of-control deviation range: When HV4 is above this range, synergistic bias forms a vicious cycle, such as persistent anxiety, high sugar intake, long-term blood sugar disorder, loss of control over appetite impulses, and continuous expansion of energy input bias.

[0249] This disclosed embodiment can locate the interactive closed loop of emotion-metabolism-eating. By using threshold ranges, it distinguishes between controllable deviations and out-of-control cycles, upgrading intervention from blood sugar control to emotion management plus low-GI substitution, such as replacing sweets with nuts to stabilize blood sugar, and mindfulness training to break the anxiety-eating closed loop.

[0250] In some embodiments, a fifth hidden variable is generated based on temporal features of movement and non-temporal features of body weight.

[0251] On the dynamic side, the temporal features of exercise are extracted: exercise intensity, pace, heart rate, and heart rate recovery time are extracted to characterize the dynamic effects of exercise intervention;

[0252] On the static side, non-time-series weight characteristics: anchored to BMI and muscle mass, reflecting the static baseline of weight management;

[0253] Synergistic mechanism: Adaptive neural networks are used to learn the adaptation patterns of baseline weight, exercise intensity, and weight changes. For example, obese users may not receive sufficient stimulation from low-intensity running, so strength training is needed to increase muscle mass.

[0254] The fifth hidden variable, also known as the exercise-weight fit synergy index, is denoted as HV5. The fifth hidden variable can be used to quantify the personalized fit between exercise intervention type and baseline weight, reflecting the synergistic effect of dynamic exercise intensity and static weight attributes.

[0255] Weight fit threshold: The minimum effective threshold for exercise-weight synergy. Values ​​below this threshold indicate that exercise intensity and weight baseline are not significantly matched, and weight changes are extremely slow or nonexistent.

[0256] Health fit range: When HV5 is in this range, exercise and weight loss are synergistically enhanced. For example, obese users can combine strength training to increase muscle mass, thereby increasing their basal metabolic rate and steadily losing weight.

[0257] Inefficient adaptation range: When HV5 is above this range, there is a co-adaptation defect. For example, obese users who run at low to moderate intensity for a long time do not gain muscle mass, have insufficient increase in basal metabolic rate, and their weight loss stagnates.

[0258] By using threshold ranges, we can distinguish between synergistic and deficient adaptation states, and upgrade the intervention to a baseline-adaptive program, such as adjusting obese users to a combination of running and strength training to accelerate metabolic weight loss.

[0259] In this embodiment, the approach moves from mapping single features to quantifying synergistic effects: it uncovers the interaction between dynamic changes and static baselines, such as the regulation of heart rate fluctuations by body constitution. Threshold ranges are used to define the health-disorder boundaries of metabolic synergy; for example, the metabolic disorder range for HV1 corresponds to the risk of fat accumulation caused by body constitution and heart rate. Each hidden variable focuses on a specific metabolic sub-dimension, such as water, energy, and sleep. A threshold system distinguishes between synergistic and imbalanced states; for example, the high-efficiency synergistic range for HV2 guides the maintenance of the exercise-diet combination, enabling intervention strategies to be precisely designed for sub-dimensions. This approach overcomes the limitations of traditional methods that only see data without considering correlations, allowing AI models to understand the synergistic operating logic of the metabolic system, the coupling patterns between dynamic processes and static baselines, and quantifying health-disorder states through threshold ranges, providing an executable technical basis for personalized interventions.

[0260] In some embodiments, such as Figure 2 As shown, S1140 may include:

[0261] S1141: Determine the body fat adjustment direction of the target user based on the fat gain / loss information;

[0262] S1142: Based on the body fat regulation direction and the metabolic profile, determine an exercise intervention strategy and / or an AI intervention strategy. The AI ​​intervention strategy uses virtual reality (VR) and / or augmented reality (AR) to intervene in the target user's appetite and / or exercise intensity.

[0263] In some embodiments, S1141: The direction of body fat regulation is deduced from the results of fat increase / decrease, such as fat loss, muscle gain, and maintenance, to clarify the core goals of intervention. S1142: Combining sub-dimensional features of metabolic profiling, such as energy metabolism stability and water metabolism efficiency, exercise intervention strategies are customized, and AI-driven appetite suppression and dynamic regulation of exercise intensity are achieved through VR and / or AR technologies, constructing a closed loop of physiological data-digital scene-behavioral intervention.

[0264] In some embodiments, fat gain / loss information—data on changes in fat mass obtained through body fat scales, DXA scans, etc., such as weekly body fat percentage and 0.3% (fat increase) and -0.2% (fat decrease)—is a direct observation indicator for weight regulation.

[0265] In some embodiments, body fat regulation is directed by weight management goals derived from changes in fat mass, combined with data such as muscle mass and water percentage. For example, fat loss requires reducing body fat percentage, while muscle gain requires increasing muscle mass while controlling fat.

[0266] In this embodiment, the direction of intervention is anchored by the quality of fat gain or loss (rather than the quantity of body weight). For example, traditional methods may mistakenly interpret unchanged body weight as metabolic stability. This embodiment identifies metabolic disorders caused by hidden energy surplus by fat gain, making the intervention more precise, and focuses on the synergy between fat breakdown and metabolism.

[0267] Step S1142: Combine body fat regulation direction and metabolic profile to determine exercise and AI pre-strategies. For example, the exercise intervention strategy is based on body fat regulation direction (fat loss / muscle gain) and metabolic profile sub-dimensions, such as energy metabolism stability and water metabolism efficiency, to customize the type, intensity, and timing of exercise. Another example is that fat loss requires moderate-intensity aerobic and strength training, while simultaneously improving energy metabolism.

[0268] AI intervention strategies (VR / AR carriers): Utilize immersive virtual reality (VR) scenarios to reshape appetite perception (e.g., a virtual refreshing environment to suppress snacking desires), or augmented reality (AR) real-time physiological feedback to regulate exercise intensity (e.g., dynamically adjusting the difficulty of virtual scenes based on heart rate), to achieve digital scene-driven behavioral changes.

[0269] In some embodiments, VR interventions target appetite, specifically energy metabolic surplus and high appetite urges.

[0270] In this scenario, metabolic profiles can reflect the target user's low energy metabolism stability (large postprandial blood glucose fluctuations → risk of insulin resistance), non-temporal dietary characteristics showing the proportion of sweets (e.g., 35% of wallpapers), and appetite impulse index, for example, the probability of binge eating when anxious is 0.8.

[0271] To address this scenario, VR intervention was introduced. During VR intervention, a low-appetite trigger scenario was constructed (such as a virtual forest hiking environment, using refreshing visual / auditory elements to suppress appetite urges). The VR scenario was automatically triggered during the user's habitual eating time (such as the 5:00 PM peak anxiety period), and the appetite index was reduced through neurofeedback (EEG monitoring of focus and adjustment of scenario elements).

[0272] After users wear VR headsets and enter a forest scene, their appetite impulse index changes from 8 to 5 (threshold 6 is a controllable range), their sweet food intake decreases, and their energy metabolism stability gradually improves, thus reducing the fluctuation range of post-meal blood sugar.

[0273] In some embodiments, AR interventions are used to increase exercise intensity in response to inefficient energy metabolism and exercise adaptation.

[0274] Metabolic profiling serves as input for intervention strategies, reflecting low energy metabolism stability in target users (e.g., low residual energy after exercise and weak afterburn effect) and exercise temporal characteristics such as running pace and heart rate recovery time. AR intervention is used for this scenario, overlaying a dynamic intensity adjustment scenario onto real-world exercise. For example, during outdoor running, AR glasses display a virtual energy monster whose speed / difficulty dynamically adjusts based on the user's heart rate and electromyographic signals. Through a physiological feedback loop (e.g., when heart rate > 150 bpm, the monster accelerates, forcibly increasing exercise intensity; when heart rate recovery is slow, the monster decelerates, prolonging the low-intensity fat-burning phase), the exercise-metabolic synergy is optimized. Thus, after AR exercise, the user experiences a prolonged afterburn effect and a weekly decrease in body fat percentage.

[0275] This disclosure, through its embodiments, combines the dynamic and static characteristics of metabolic profiling (such as energy metabolism stability and exercise intensity adaptability) to upgrade exercise programs from general templates to metabolically synergistic programs. For example, to address inefficient energy metabolism, it designs a combination of AR dynamic intensity training and strength training. The R / AR intervention breakthrough: traditional interventions rely on willpower (such as self-controlled blood sugar and consistent exercise), while this solution reshapes behavioral triggering conditions through digital scenarios, using VR to suppress appetite urges and AR to enhance exercise intensity, thus lowering the intervention execution threshold and improving adherence.

[0276] In summary, the approach shifts from outcome-based intervention to cause-driven intervention: fat gain / loss information anchors to metabolic imbalances, while metabolic profiling breaks down the causes of these imbalances, such as low energy metabolism stability and insulin resistance. This upgrades intervention strategies from blindly controlling weight to precisely addressing underlying causes, such as using VR to suppress high-GI eating impulses and reduce energy surplus at its source. From physical intervention to digital collaboration: VR / AR technology transforms physiological data into digital scene commands (such as heart rate and virtual monster speed), constructing a real-time closed loop of physiological state, digital environment, and behavioral change. Compared to the delayed feedback of traditional exercise tracking and diet records, this increases intervention response speed by 50% (e.g., AR adjusts exercise intensity in real time to avoid over- or under-exercising). From single-dimensional to multi-modal collaboration: Exercise strategies and AI interventions (VR / AR) work synergistically, covering both exercise expenditure and behavioral inhibition dimensions. For example, AR improves fat-burning efficiency, while VR reduces energy intake.

[0277] This embodiment, through in-depth collaboration of metabolic profiling and behavioral reshaping using VR / AR technology, upgrades weight regulation from passive execution to proactive digital intervention that adapts to metabolic patterns, thereby achieving precise health management.

[0278] In some embodiments, the method further includes at least one of the following:

[0279] When the system detects that the target user is eating or has an appetite urge, it controls the head-mounted device to output perceptual interference information according to the AI ​​intervention strategy; the perceptual interference information includes: visual interference information, tactile interference information and / or memory interference information;

[0280] The perceptual interference information is used to intervene in the target user's perception, in order to suppress or enhance appetite;

[0281] When the target user exercises, exercise interference information is output based on the target user's metabolic model and body fat regulation direction; the exercise interference information is used to strengthen or weaken the exercise intensity of the target user.

[0282] In this embodiment of the disclosure, a closed loop is constructed, from physiological state and digital scene to behavioral intervention, and two core types of intervention are achieved through head-mounted devices (VR / AR):

[0283] Eating / Appetite Intervention: Based on metabolic profiling, identify high-risk appetite characteristics (such as anxious eating, high-GI diet dependence), output visual, tactile, and memory interference information, and regulate appetite perception in multiple dimensions (inhibit excessive intake or enhance healthy eating).

[0284] Exercise intervention: Combining the real-time status of the metabolic model (energy metabolism efficiency, exercise afterburn effect) with the direction of body fat regulation (fat loss or muscle gain), dynamically output exercise interference information (virtual resistance or scenario incentives) to precisely strengthen or weaken exercise intensity and optimize metabolic synergy.

[0285] Interventions targeting sensory disturbances during eating / appetite can be based on multi-sensory reshaping of appetite triggers. Metabolic profiling can be used to uncover the correlation between appetite and metabolic quality control (e.g., anxiety, high-GI diet, blood sugar disorders, leading to more anxious eating). Multimodal stimulation using VR / AR can break the vicious cycle of metabolic disorders caused by unhealthy appetite triggers.

[0286] For example, for emotional eating, visual distractions reduce the appeal of high-GI foods.

[0287] Assume that the target's metabolic profile reflects the target user's appetite impulse index (e.g., appetite when anxious), non-temporal dietary characteristics (proportion of sweets), and blood glucose fluctuation.

[0288] AI intervention trigger: Identify the target user's peak anxiety time (17:00) and habitual sweets scene (cake on the desk).

[0289] VR interventions may include:

[0290] Visual interference: AR headsets overlay low-attractive filters onto sweets (such as cakes) in the physical environment, such as reducing color saturation and increasing blurring effects, to weaken visual stimulation.

[0291] Environmental replacement: The virtual background automatically switches to a forest and stream scene, using natural elements to replace visual cues of food and suppress appetite urges. In this way, the target user's desire for sweets shifts from a strong craving to a lack of obvious need, resulting in reduced actual sweets intake, decreased blood sugar fluctuations, and thus bringing metabolic deviations into a controllable range.

[0292] Tactile interference targets rapid eating habits, simulating a feeling of fullness to inhibit overeating.

[0293] Metabolic profiles can reflect the stability of energy metabolism and the timing characteristics of eating (e.g., duration of eating).

[0294] AI intervention triggered: Identifying the target user's mealtime.

[0295] AR interventions may include:

[0296] Pressure feedback: When chewing the 10th bite, the head-mounted display releases a slight pressure pulse in the temporal region to simulate the feeling of stomach expansion and enhance the sensation of having eaten.

[0297] Vibration alert: When the calorie intake exceeds 200kcal, the vibration frequency of the head-mounted display increases with the increase in calories. This increases the target user's meal time, allows for more thorough chewing, promotes the secretion of digestive hormones, reduces calorie intake per meal, and brings energy metabolism stability into a healthy and coordinated range.

[0298] In some embodiments, memory interference is introduced to reinforce healthy eating preferences in response to metabolic deviation cycles. In some embodiments, metabolic profiling reflects eating motivation and the metabolic deviation co-index HV4.

[0299] Intervention trigger: Identify the target user's appetite urge, for example, the afternoon tea time at 3:00 PM.

[0300] AR interventions may include:

[0301] Visual memory recall: Virtually displays photos of last week's healthy meals and stable blood sugar curves to reinforce the association between healthy eating and metabolic stability. Voice feedback trigger: Automatically plays back pleasant experiences after past healthy eating, such as the memory of feeling energetic after eating a salad.

[0302] In some embodiments, the metabolic profile includes: metabolically sensitive and metabolically insensitive types based on nutrient absorption conversion criteria; the metabolic profile further includes: based on the sensory interference information, including at least one of the following:

[0303] A food zoom-in view adapted to the metabolic profile and the body fat regulation direction;

[0304] Tactile interference with the stickiness of food gripping, which is compatible with the metabolic profile and the direction of body fat regulation;

[0305] The system uses visuals of blood sugar fluctuations or energy alerts that are adapted to the metabolic profile and the direction of body fat regulation in order to interfere with the user's memory.

[0306] In some embodiments, this disclosure uses metabolic profiling as a core classification method, combined with body fat regulation direction, to output perceptual interference information that is precisely adapted to individual metabolic characteristics through head-mounted devices (VR / AR), achieving three major intervention dimensions:

[0307] Visual interference: Dynamically zoom in and out on food images to enhance or diminish the appeal of different GI foods to different user groups;

[0308] Tactile interference: Simulates the difference in stickiness when grasping food to regulate eating desire and rhythm;

[0309] Memory interference: Overlaying blood glucose fluctuation animations or energy cues reshapes the memory associated with diet and metabolism.

[0310] Thus, metabolic typing determines the intensity and form of interference. Achieving metabolic sensitivity requires fine-tuning to avoid metabolic adaptation, while achieving acute sensitivity requires strong stimulation to disrupt homeostasis, so that the intervention effect closely matches the individual's metabolic response pattern.

[0311] The table below shows the interference relationships between the core characteristics of metabolically sensitive and metabolically acute types.

[0312]

[0313] For those with metabolic sensitivity, the applicability of food is precisely adjusted through subtle zooming; for those with metabolic sluggishness, extreme zooming and filters are used to forcefully guide healthy eating choices.

[0314] Assuming a metabolically sensitive type and a fat loss orientation, and given that the metabolic profile indicates a metabolically sensitive type, the body fat regulation direction is fat loss, requiring the suppression of high-GI food intake to stabilize energy metabolism.

[0315] The visual interference trigger condition is that when the target user looks at high-GI foods (such as cakes and white bread), the AR headset automatically initiates intervention. The image of high-GI foods is dynamically reduced in size to decrease visual impact, leveraging the user's sensitivity to visual cues to weaken the desire to eat; the image of low-GI foods (such as oats and brown rice) is enlarged to emphasize the visual priority of healthy foods and guide active selection.

[0316] Assuming a metabolically sensitive and weight-loss-oriented user, the visual interference trigger condition is as follows: When the target user looks at high-GI foods (such as French fries and milk tea), the AR headset automatically activates an extreme weakening strategy. Specifically, this includes: shrinking the high-GI food image to 50% and overlaying an "oily" filter to drastically reduce visual appeal, leveraging the user's need for strong stimuli to respond, thus forcibly weakening the choice of unhealthy foods. Conversely, enlarging the high-GI food image (such as broccoli and buckwheat noodles) to 150% and overlaying a "fresh" filter to strengthen the visual anchor of healthy foods and establish a positive association for selection.

[0317] Tactile interference adapts the stickiness of food to different metabolic types. For those with metabolic sensitivity, subtle tactile feedback optimizes eating rhythm and nutrient selection; for those with metabolic urgency, strong tactile pressure forces changes in eating habits.

[0318] For metabolically sensitive individuals and those aiming to build muscle, haptic feedback is used to optimize protein intake. Tactile interference trigger conditions: When a user grasps high-protein foods (such as chicken breast or steak) or high-carbohydrate foods (such as white bread or cake), the AR controller activates a distorted haptic strategy. Specifically, this may include: when grasping high-protein foods, the controller simulates a light and elastic feel to enhance the sense of health and easy absorption, leveraging the sensitive haptic response to improve eating pleasure and encourage proactive choices. When grasping high-carbohydrate foods, the controller simulates a sticky and resistive feel to reduce the desire for unnecessary carbohydrate intake, guiding a reduction in low-value energy intake.

[0319] Assuming a metabolically stimulating and muscle-building orientation, this design uses strong tactile pressure to force dietary adjustments. The tactile interference trigger condition is as follows: When the target user grasps high-protein foods (such as fish, protein powder) or low-protein foods (such as fries, cake), the AR controller activates a strong pressure tactile strategy. Specifically, when grasping high-protein foods, the controller simulates a heavy, firm feel to enhance the sensation of high protein and high energy, leveraging the characteristic of metabolic stimulation requiring strong stimulation to forcibly establish a healthy dietary association. When grasping low-protein foods, the controller simulates a loose, resistive feel to drastically weaken the appeal of non-muscle-building foods, reducing the willingness to choose them.

[0320] Memory interference is used to adapt blood glucose fluctuations / energy cues to metabolic typing. For metabolically sensitive types, dynamic blood glucose curves are used to strengthen the real-time association between diet and metabolism; for metabolically sensitive types, historical energy comparisons are used to forcibly reshape long-term eating habits.

[0321] For those with metabolic sensitivity and those aiming for weight loss, dynamic blood glucose curves enhance memory. Memory interference trigger conditions: When the target user selects a low-GI or high-GI food, the AR headset automatically overlays metabolic-related prompts. Specifically, this may include: after selecting low-GI foods (such as oats and brown rice), overlaying a stable blood glucose curve and energy consumption prompts, leveraging the rapid memory reconstruction characteristic of sensitive users to strengthen the positive association between healthy eating and metabolism. After selecting high-GI foods (such as fried dough sticks and milk tea), overlaying a blood glucose spike curve and energy accumulation prompts, strengthening the negative metabolic association between unhealthy eating and metabolism.

[0322] For those seeking a metabolic boost and those aiming for weight loss, historical energy levels are used to forcefully reshape their memory. Memory interference trigger conditions: When the target user selects a high-GI or low-GI food, the AR headset automatically displays a historical energy trend comparison. Specifically, selecting high-GI foods (such as pizza or cake) triggers a warning about accumulated historical energy levels, leveraging the need for a strong memory impact to forcibly disrupt old, stable perceptions. Selecting low-GI foods (such as broccoli or brown rice) triggers a historical energy expenditure incentive, reinforcing positive memories of healthy eating.

[0323] In summary, the precise adaptation of metabolic subtyping in this embodiment is achieved by designing fine-tuning and strong stimulation strategies based on the differences in metabolic responses between metabolically sensitive and metabolically acute subtyping, thus improving the intervention effect. Furthermore, the synergistic effect of multimodal interference—the combined effects of vision (image scaling), touch (grasping stickiness), and memory (blood glucose / energy cues)—covers the entire chain of sensory triggering, behavioral selection, and memory reshaping, resulting in a higher success rate of appetite regulation compared to single-modal intervention. Active shaping of metabolic homeostasis is also achieved: for metabolically sensitive subtyping, subtle interference maintains dynamic metabolic balance (avoiding overstimulation that triggers metabolic adaptation); for metabolically acute subtyping, strong interference forcibly disrupts the old homeostasis (initiating metabolic adjustment), increasing the achievement rate of weight regulation direction.

[0324] This embodiment, through the deep coupling of metabolic typing and multimodal sensory interference, upgrades AI intervention from general scenario coverage to precise scenario customization tailored to individual metabolic characteristics, providing an innovative technical path for digital health management driven by typing and multi-sensory collaboration.

[0325] In some embodiments, the motion interference information is used for virtual motion environment resistance; and / or, the motion interference information is used to strengthen or weaken motion intensity feedback.

[0326] This disclosure utilizes VR / AR to construct a virtual motion environment, and outputs two types of motion interference information based on the differentiated response characteristics of metabolically sensitive and metabolically insensitive individuals, thereby achieving precise regulation of the synergy between exercise intensity and metabolism.

[0327] Dynamic adjustment of virtual environment resistance: Based on metabolic profile, the resistance of the virtual scene (such as headwind, slope, virtual heavy objects) is changed to adapt to individual metabolic response.

[0328] Exercise intensity feedback enhancement / weakening: Through multimodal feedback including visual (muscle growth animation) and tactile (handle vibration, pressure feedback), precise control of exercise perception is achieved (e.g., enhancing muscle-building force feedback and weakening fat-burning fatigue perception). Metabolic type determines the interference mode (sensitive types require fine dynamic adjustments to maintain metabolic activity, while those with a tingling sensation require strong stimulation to disrupt homeostasis), ensuring that exercise intervention deeply aligns with individual metabolic patterns.

[0329] When users run outdoors, AR glasses construct a dynamic virtual running track, collecting heart rate and electromyography (EMG) signals in real time to adjust resistance. During indoor cycling, VR headsets create sudden resistance scenarios, randomly inserting high-resistance challenges into three weekly workout sessions. During regular cycling, random uphill resistance bursts are triggered, utilizing the sudden, strong stimulus response to forcibly activate fat metabolism pathways. After high resistance, a low-resistance recovery scenario is switched to prevent excessive fatigue and interruption of exercise. During strength training (squats, bench presses), AR controllers, combined with EMG sensors, output dynamic tactile feedback.

[0330] If muscle activation is less than the activation threshold of 1 (insufficient force), the handle will simulate slight vibration and gradually increase pressure (strengthening requires increasing the sense of force exertion and optimizing the quality of movement).

[0331] If muscle activation exceeds the activation threshold of 2 (excessive force), the simulated vibration of the handle will decrease to reduce the pressure (to weaken the impulse to continue applying force and avoid injury).

[0332] During strength training (deadlifts, pull-ups), the VR headset provides strong visual and tactile feedback. After completing the required number of sets (3 sets x 8 reps), a muscle growth animation and strong vibration of the controllers appear to enhance the connection between the exercise and muscle synthesis, breaking the growth stagnation. During the recovery phase after exercise, metabolic repair prompts are added to strengthen the connection between nutrition and muscle growth.

[0333] In some embodiments, the hierarchical coding of two body fat information focuses on the differences in the types of fat storage, solving the problem that overall body fat percentage cannot distinguish metabolic risk;

[0334] By embedding nutritional information into the composition of energy intake, dietary descriptions can be effectively linked to fat metabolism. This combination provides AI models with a dual perspective: a stored view of fat gain / loss and an input-driven view, enabling subsequent intervention strategies to better align with the specific mechanisms of individual fat metabolism.

[0335] The following combines virtual reality, VR, augmented reality, and AR technologies to extend the discussion from two aspects: the dynamic generation logic of intervention strategies and the specific execution content. The core is to achieve weight intervention by reshaping real physiological feedback through virtual perception. The technical logic is closely related to the human neuro-metabolic connection mechanism:

[0336] By constructing virtual scenarios inversely synchronized with the user's metabolic rhythm using VR, and leveraging visual-vestibular cross-stimulation to reshape the brain's perception of energy balance, this drives AI models to generate revolutionary intervention strategies. The counterintuitive logic is that traditional interventions alter metabolism through real-world behavior, while this approach distorts metabolic judgment through virtual perception, causing the brain to proactively lower its energy demands. For example, continuous weightlessness in the virtual environment suppresses the appetite center.

[0337] VR scene dynamic generation: The AI ​​model generates a metabolic reverse scene based on real-time metabolic data from wearable devices, such as blood sugar fluctuations and cortisol levels. When it detects a sudden rise in the target user's blood sugar and the impending feeling of hunger, the VR instantly switches to a hypergravity environment, where objects appear heavier and movements become slower. Through vestibular-visual linkage, it activates the illusion of sufficient energy reserves in the brain, suppressing appetite signals.

[0338] The intervention strategy is generated based on the following: The model extracts user interaction data in the VR scene, such as the duration of stay in the virtual hypergravity environment and the frequency of avoiding high-energy virtual food. This data is then integrated with real metabolic data to generate virtual anchoring strength parameters. For example, the hypergravity scene needs to last for 15 minutes to counteract the urge to consume a meal of calories. These parameters serve as the core variables of the intervention strategy.

[0339] AR can be used to manipulate users' perception of their own behavior in real time, such as mapping real eating actions to virtual overeating, and light exercise to virtual high-intensity exercise, thereby driving real-world behavioral correction through cognitive dissonance. The counterintuitive logic is that instead of directly asking users to eat less and exercise more, AR creates cognitive biases that encourage users to actively avoid real-world overeating, such as stopping after seeing themselves eat three bowls of rice in AR when they actually only eat one.

[0340] AR-based food mirroring intervention: Users wear AR glasses while eating. The AR magnifies the visual size of each bite of food by 2-3 times. For example, if a user actually eats one bite of cake, the AR will display it as eating one slice of cake. Simultaneously, a virtual calorie counter is displayed, creating a real-time, artificially accumulated value. For instance, if the user actually ingests 200 kcal, the AR will display it as having ingested 500 kcal. This leverages the brain's preferential trust in visual information to trigger a false sense of fullness, reducing actual food intake.

[0341] Augmented AR Motion Perception: When users engage in light activities, such as walking, AR maps their movements to high-intensity exercise in a virtual scene. For example, while walking, AR displays that the user is hiking, with the stride frequency corresponding to the virtual hiking speed. Simultaneously, virtual motion sound effects such as rapid breathing and heartbeat are played through bone conduction headphones. Based on the principle of perception-physiology linkage, the virtual high-intensity signal stimulates the sympathetic nervous system, increasing the actual metabolic rate. Visual cues can increase energy consumption by 15%-20% for the same amount of exercise.

[0342] VR / AR cross-modal metabolic memory rewriting closed-loop intervention includes:

[0343] By constructing a virtual identity representing an ideal metabolic state using VR, such as a 3D avatar of a user's ideal weight, and combining this with AR to overlay the behavioral feedback of the virtual identity onto the real world, users can reconstruct their real metabolic habits by immersing themselves in virtual memories. VR metabolic memory implantation: Every night, users wear VR devices to enter an ideal life scenario. The virtual avatar acts according to a healthy mode. For example, when the virtual avatar selects a vegetable salad, the user uses VR controllers to experience the texture of the virtual food, while the system releases trace amounts of odor molecules corresponding to the food. Through multi-sensory stimulation, healthy behavioral memories are formed in the hippocampus of the brain.

[0344] AR Memory Anchoring Execution: During the day, when users are in a real-world environment, AR automatically recognizes the scene, such as passing by a convenience store, and instantly overlays decision prompts from a virtual avatar. For example, the virtual avatar shakes its head to refuse potato chips in AR, and displays a message saying, "You said this in VR last week, it will make you tired," using virtual memory to suppress real impulses. The AI ​​model dynamically adjusts the intensity of sensory stimulation in the VR scene by comparing the deviation between virtual memory and real behavior, such as the difference between virtual preferences and real choices. The greater the deviation, the more realistic the touch / smell of the virtual scene.

[0345] In some embodiments, a VR-based positive metabolic anchoring strategy is employed for fat gain. Based on latent variable extraction, distinguishing between changes in fat and water and metabolic profiling technology, the VR scene design in the fat gain scenario forms an inverse logic with weight loss. When the wearable device detects that the user's metabolic rate is too high, such as a basal metabolic rate >1800kcal / day, which is not conducive to calorie storage, VR generates a microgravity virtual environment. Objects appear lighter and movements are lighter, transmitting a low energy consumption signal to the brain through vestibular-visual linkage, stimulating the hypothalamus's feeding center. Experiments show that microgravity visual cues can increase the secretion of the appetite-related hormone ghrelin by 20%-30%.

[0346] For individuals with poor digestion and absorption who are prone to weight gain, the metabolic profile is labeled with low nutrient utilization efficiency. VR is used to create a virtual scenario of efficient absorption: when users wear VR devices, their virtual avatars will display the dynamic effect of rapid nutrient absorption in real time after eating, such as the animation of the concentration of nutrients rising in the virtual blood. This visual feedback reinforces the psychological suggestion that eating is effective and increases the willingness to eat actively.

[0347] AR-guided behavioral perception enhancement intervention for fat gain includes: relying on multi-source signal fusion to integrate data such as weight and metabolism, AR promotes calorie surplus by weakening the negative feedback of overeating and the perception of exercise expenditure. In the fat gain scenario, the AR eating mirror is designed in the opposite way to weight loss: the visual size of real food is reduced, such as eating 1 bite of steak in reality, but it is displayed as eating 1 / 3 bite in AR. At the same time, the virtual calorie count bar underestimates the intake, so when the actual intake is 300kcal, it is displayed as only 150kcal, reducing the brain's alertness to overeating and prolonging the eating time.

[0348] For fat gainers who need to reduce unnecessary energy consumption, such as those with excessive daily activity levels in their metabolic profile, AR maps real light activities, such as walking, into virtual low-intensity scenarios. For example, what appears as walking on flat ground in AR may actually be climbing a hill. At the same time, it weakens the sound effects of breathing, heartbeat, and other movements. By reducing the perceived amount of exercise, AR reduces the actual amount of activity. Data shows that this kind of perception adjustment can reduce the average daily energy consumption by 10%-15%.

[0349] Metabolic memory rewriting adapted for fat gain can include: based on a closed-loop feedback mechanism, VR / AR reshapes fat gain habits by implanting positive memories of calorie reserves. In VR, an ideal weight virtual avatar is constructed, corresponding to the healthy body shape of the fat gain goal. The virtual avatar's eating scenarios reinforce the association between high-calorie foods and pleasure. For example, when the virtual avatar eats nuts, the VR device releases the corresponding aroma, while the virtual environment switches to a comfortable resting scene. Through multi-sensory stimulation, a memory anchor is formed that high-calorie eating equals pleasure.

[0350] AR activates fat-gain memories in real-world scenarios: When users hesitate about whether to have an extra meal, AR automatically overlays encouraging prompts from a virtual avatar, such as if you said in VR last week that this energy bar could help you reach your goal. Combined with real-time weight data from wearable devices, such as if you have gained 0.1kg since yesterday, it reinforces the positive feedback that persistence is effective.

[0351] like Figure 3 As shown, this disclosure provides an artificial intelligence (AI) data analysis device based on a wearable device, including:

[0352] The acquisition module 3110 is used to acquire multi-source heterogeneous data of the target user, wherein the multi-source heterogeneous data includes at least data acquired through wearable devices;

[0353] Extraction module 3120 is used to process the multi-source heterogeneous data through an AI model to extract hidden variables;

[0354] The acquisition module 3130 is used to acquire a metabolic profile of the target user based on the hidden variables; the metabolic profile is used to reflect the target user's water metabolism and energy metabolism.

[0355] Strategy module 3140 is used to generate intervention strategies based on the target user's metabolic profile and fat gain / loss information;

[0356] Intervention module 3150 is used to perform intervention operations according to the intervention strategy.

[0357] The AI ​​intervention device may include one or more electronic devices, which can, for example... Figure 4 As shown. This AI intervention device can realize the artificial intelligence (AI) data analysis method based on wearable devices provided by any of the aforementioned technical solutions.

[0358] Figure 4 As shown, this application embodiment provides an electronic device including a processor 10 and a memory 11. Optionally, the device may further include a communication interface 12 and a bus 9. The processor 10, communication interface 12, and memory 11 can communicate with each other via the bus 9. The communication interface 12 can be used for information transmission. The processor 10 can call logical instructions in the memory 11 to execute the queue-based voiceprint data processing method of the above embodiment.

[0359] Furthermore, the logical instructions in the aforementioned memory 11 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0360] The memory 11, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as program instructions / modules corresponding to the methods in the embodiments of this application. The processor 10 executes functional applications and data processing by running the program instructions / modules stored in the memory 11, thereby realizing the AI ​​data analysis method based on wearable devices in the above embodiments.

[0361] The memory 11 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created based on the use of the electronic device. Furthermore, the memory 11 may include high-speed random access memory and may also include non-volatile memory.

[0362] This application provides a computer program product, which includes a computer program stored on a storage medium. The computer program includes program instructions, which, when executed by a computer, cause the computer to perform the aforementioned artificial intelligence (AI) data analysis method based on wearable devices.

[0363] The aforementioned computer-readable storage medium may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.

[0364] The technical solutions of this application embodiment can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions to cause a computer device, such as a personal computer, server, or network device, to execute all or part of the steps of the method described in this application embodiment. The aforementioned storage medium can be a non-transitory storage medium, including various media capable of storing program code such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks, or it can be a transient storage medium.

[0365] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0366] The embodiments or examples disclosed in this application are not exhaustive, but merely illustrative of some embodiments or examples, and are not intended to limit the scope of protection of this disclosure. Unless contradictory, each step in a particular embodiment or example can be implemented as an independent embodiment, and the steps can be arbitrarily combined. For example, a solution after removing some steps in a particular embodiment or example can also be implemented as an independent embodiment, and the order of the steps in a particular embodiment or example can be arbitrarily interchanged. Furthermore, optional methods or examples in a particular embodiment or example can be arbitrarily combined; moreover, embodiments or examples can be arbitrarily combined. For example, some or all steps of different embodiments or examples can be arbitrarily combined, and a particular embodiment or example can be arbitrarily combined with optional methods or examples of other embodiments or examples.

[0367] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0368] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0369] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0370] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device, such as a personal computer, server, or network device, to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory, ROM, random access memory, RAM, portable hard drives, magnetic disks, or optical disks.

[0371] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for artificial intelligence (AI) data analysis based on wearable devices, characterized in that, include: Collect multi-source heterogeneous data of the target user, wherein the multi-source heterogeneous data includes at least data collected through wearable devices; The multi-source heterogeneous data is processed using an AI model to extract latent variables. This process includes: standardizing the multi-source heterogeneous data to obtain a three-dimensional data matrix; the three-dimensional data matrix includes a first matrix and a second matrix; the first matrix includes time-series data extracted from the multi-source heterogeneous data; the second matrix includes non-time-series data extracted from the multi-source heterogeneous data; time-series features are extracted from the first matrix based on a time correlation algorithm; these time-series features reflect the dynamic physiological processes of the target user; and non-time-series features are extracted from the second matrix based on hot coding or embedding layer transformation; these non-time-series features reflect the dynamic physiological processes of the target user. The static physiological state; based on the synergistic effect of dynamic physiological processes and static physiological states, the hidden variables are obtained by fusing the temporal features and the non-temporal features; wherein, the process of obtaining the hidden variables by fusing the temporal features and the non-temporal features based on the synergistic effect of dynamic physiological processes and static physiological states includes at least: fusing temporal features of heart rate and non-temporal features of body fat to obtain a first hidden variable; the first hidden variable is used to determine whether the target user is in a healthy synergistic range or a metabolic disorder range; when the first hidden variable is in the healthy synergistic range, the target user's heart rate fluctuations and body type have a benign synergistic effect; when the first hidden variable is in the metabolic disorder range, the target user has synergistic metabolic damage in water metabolism and fat metabolism; Based on the hidden variables, a metabolic profile of the target user is obtained; the metabolic profile is used to reflect the target user's water metabolism and energy metabolism; the metabolic profile is either metabolically sensitive or metabolically insensitive; the metabolically sensitive type responds to intervention faster than the metabolically insensitive type, and the metabolically sensitive type is more likely to plateau in its intervention response compared to the metabolically insensitive type. Based on the target user's metabolic profile and fat gain / loss information, an intervention strategy is generated; the intervention strategy includes at least hydration designed for the metabolic disorder zone or a diet that takes glycemic index into account; The intervention operation is performed according to the intervention strategy.

2. The method according to claim 1, characterized in that, The multi-source heterogeneous data of the target user collected includes at least one of the following: The weighing device acquires the target user's weight and / or body fat information. The device acquires at least one of the target user's heart rate information, activity level information, and sleep information collected by the limb-worn device. Obtain the metabolic information of the target user collected by the blood glucose meter; Use head-mounted devices to collect appetite information and / or emotional state; The target user's dietary information is obtained through diet-related applications.

3. The method according to claim 1, characterized in that, The hidden variable obtained by fusing the temporal features and the non-temporal features based on the synergistic effect of dynamic physiological processes and static physiological states also includes at least one of the following: By integrating temporal features of exercise volume and non-temporal features of diet, a second hidden variable is obtained. This second hidden variable is used to express the synergistic relationship between the target user's dynamic energy metabolism, the post-exercise fat consumption effect, and dietary capacity intake. The second hidden variable is used to determine whether the target user is in a highly efficient synergistic zone or a poorly efficient synergistic zone. When the target user is in the highly efficient synergistic zone, the synergy between exercise and diet is enhanced; when the target user is in the poorly efficient synergistic zone, the synergy between exercise and diet is ineffective. By integrating sleep temporal features and emotion non-temporal features, a third hidden variable is obtained; the third hidden variable is used to reflect the sleep-emotion co-metabolic effect; the third hidden variable is used to determine whether the target user is in a steady-state co-metabolic range or an injury co-metabolic range; when the target user is in the steady-state co-metabolic range, the target user's sleep and emotion co-metabolic effects are stable; when the target user is in the injury co-metabolic range, the target user's sleep and emotion co-induce metabolic damage. By integrating temporal metabolic features and non-temporal dietary features, a fourth hidden variable is obtained. This fourth hidden variable is used to reflect dietary motivation, metabolic status, and appetite-driven energy input deviation information. It is also used to determine whether the target user is in a controllable deviation range or an uncontrollable deviation range. When the target user is in the controllable deviation range, their dietary and metabolic deviations can be regulated by the metabolic system. When the target user is in the uncontrollable deviation range, their dietary and metabolic deviations enter a vicious cycle. By fusing temporal features of exercise and non-temporal features of body weight, a fifth hidden variable is obtained; The fifth hidden variable is used to reflect the personalized adaptation of exercise intervention and weight baseline; the fifth hidden variable is used to determine whether the target user is in the healthy adaptation range or the inefficient adaptation range; when the target user is in the healthy adaptation range, exercise and weight synergistically enhance each other; when the target user is in the inefficient adaptation range, the synergistic adaptation of exercise and weight for the target user fails.

4. The method according to claim 1 or 2, characterized in that, The step of generating an intervention strategy based on the target user's metabolic profile and fat gain / loss information includes: The body fat adjustment direction for the target user is determined based on the fat gain / loss information. Based on the body fat regulation direction and the metabolic profile, an exercise intervention strategy and / or an AI intervention strategy are determined; wherein, the AI ​​intervention strategy uses virtual reality (VR) and / or augmented reality (AR) to intervene in the target user's appetite and / or exercise intensity.

5. The method according to claim 4, characterized in that, The method further includes at least one of the following: When the system detects that the target user is eating or has an appetite urge, it controls the head-mounted device to output perceptual interference information according to the AI ​​intervention strategy; the perceptual interference information includes: visual interference information, tactile interference information and / or memory interference information; The perceptual interference information is used to intervene in the target user's perception, in order to suppress or enhance appetite; When the target user exercises, exercise interference information is output based on the target user's metabolic model and body fat regulation direction; the exercise interference information is used to strengthen or weaken the exercise intensity of the target user.

6. The method according to claim 5, characterized in that, The metabolic profile includes: metabolically sensitive and metabolically insensitive types based on nutrient absorption conversion criteria; the metabolic profile also includes: based on the sensory interference information, including at least one of the following: A food zoom-in view adapted to the metabolic profile and the body fat regulation direction; Tactile interference with the stickiness of food gripping, which is compatible with the metabolic profile and the direction of body fat regulation; The system uses visuals of blood sugar fluctuations or energy alerts that are adapted to the metabolic profile and the direction of body fat regulation in order to interfere with the user's memory.

7. The method according to claim 5, characterized in that, The motion interference information is used for virtual motion environment resistance; and / or, the motion interference information is used to strengthen or weaken motion intensity feedback.

8. An artificial intelligence (AI) data analysis device based on wearable devices, characterized in that, include: The acquisition module is used to acquire multi-source heterogeneous data of the target user, wherein the multi-source heterogeneous data includes at least data acquired through wearable devices; An extraction module is used to process the multi-source heterogeneous data using an AI model to extract hidden variables. Specifically, the extraction module performs data standardization on the multi-source heterogeneous data to obtain a three-dimensional data matrix. The three-dimensional data matrix includes a first matrix and a second matrix. The first matrix includes time-series data extracted from the multi-source heterogeneous data. The second matrix includes non-time-series data extracted from the multi-source heterogeneous data. Time-series features are extracted from the first matrix based on a time correlation algorithm. These time-series features reflect the dynamic physiological processes of the target user. Non-time-series features are extracted from the second matrix based on hot coding or embedding layer transformation. The non-time-series features are used to reflect the static physiological state of the target user. Based on the synergistic effect of dynamic physiological processes and static physiological states, the time-series features and the non-time-series features are fused to obtain the hidden variable, which includes at least: fusing heart rate time-series features and body fat non-time-series features to obtain a first hidden variable; the first hidden variable is used to determine whether the target user is in a healthy synergistic range or a metabolic disorder range; when the first hidden variable is in the healthy synergistic range, the target user's heart rate fluctuations and body type have a benign synergistic effect; when the first hidden variable is in the metabolic disorder range, the target user has synergistic metabolic damage in water metabolism and fat metabolism. The acquisition module is used to acquire the metabolic profile of the target user based on the hidden variables; the metabolic profile is used to reflect the water metabolism and energy metabolism of the target user; the metabolic profile is either metabolically sensitive or metabolically insensitive; the metabolically sensitive type responds to intervention faster than the metabolically insensitive type, and the metabolically sensitive type is more likely to plateau in its intervention response compared to the metabolically insensitive type. The strategy module is used to generate intervention strategies based on the target user's metabolic profile and fat gain / loss information; the intervention strategies include at least hydration designed for the metabolic disorder intervals or diets that take glycemic index into account. An intervention module is used to perform intervention operations according to the intervention strategy.

9. A computer-readable medium, characterized in that, It stores instructions that, when executed by one or more processors, cause the processors to perform the method as described in any one of claims 1 to 7.

Images