Personalized adult nutrition health care product customization system and method

Abstract

The invention relates to the technical field of nutrition health care product customization, and discloses a personalized adult nutrition health care product customization system and method. The method comprises the following steps: collecting biological characteristics, dietary habits, health targets and gene detection data of a user, and constructing a nutritional requirement database through labeling and preprocessing; a multi-dimensional nutrition demand tensor is generated through principal component analysis, clustering analysis and the like, the matching degree of a target user and a representative sample is calculated, a weighted matching degree vector is obtained through neural network optimization, and then a personalized nutrition formula substrate is generated and optimized through a genetic algorithm. Kalman filtering is used for adjusting a formula according to real-time health data of a user, monitoring deviation to trigger reconstruction, associating a historical scheme to update a nutrition rule base, and finally outputting a nutrition formula coded as a production instruction, so that accurate personalized customization is realized, and nutrition is safe, reasonable and traceable.

Description

Technical Field

[0001] The present invention relates to the technical field of customized nutritional health products, and specifically to a personalized adult nutritional health product customization system and method. Background Art

[0002] With the continuous improvement of people's health awareness, the demand for nutritional health products is increasing day by day. In the modern fast-paced lifestyle, adults are facing various health challenges, such as obesity, increased risk of chronic diseases, and decreased immunity, which makes their need for personalized nutrition and health care more urgent. However, most of the nutritional health products on the market are general-purpose products and cannot meet the needs of individual differences.

[0003] From the perspective of data acquisition, the research and development and recommendation of traditional nutritional health products mainly rely on general nutritional knowledge and population average data. This approach ignores the differences among individuals in terms of biological characteristics, eating habits, health goals, and genetic levels. For example, people with different ages, body fat percentages, and metabolic rates have very different nutritional needs. The elderly may need more calcium and vitamin D to maintain bone health, while young people may be more concerned about the intake of protein and vitamins to support exercise and daily activities. During the fat loss process, people with a high body fat percentage need to strictly control calorie and fat intake, while those with a low body fat percentage focus more on muscle gain and nutritional supplements to improve metabolism. At the same time, the diversity of eating habits also determines that each person's nutritional intake pattern is different. Some people may mainly be vegetarian and are prone to lack certain nutrients, while meat eaters may face other problems of nutritional imbalance. Although the development of gene detection technology has provided new possibilities for personalized nutrition, it has not been effectively integrated into the customization process of nutritional health products.

[0004] In terms of formulating nutritional plans, the existing methods lack accuracy and a dynamic adjustment mechanism. General nutritional plans often adopt a one-size-fits-all approach and cannot be optimized according to an individual's real-time health status. Taking the fat loss population as an example, during the fat loss process, as the weight decreases and the amount of exercise changes, the body's nutritional needs are also constantly changing. If the nutritional plan cannot be adjusted in a timely manner, it may lead to poor fat loss results, or even problems of nutritional deficiency or over-intake. Moreover, traditional plans rarely consider the interactions and taboos among nutritional components. The simultaneous intake of some nutritional components may have adverse effects. For example, the absorption of calcium and iron in the gastrointestinal tract interferes with each other, but the existing recommendations for nutritional health products rarely avoid this.

[0005] From a technical implementation perspective, the traditional technical means for customizing nutritional health products are relatively backward. The data processing methods are simple and cannot effectively integrate and analyze a large amount of complex multi-source data. In key links such as cluster analysis and matching degree calculation, there is a lack of advanced algorithm support, making it difficult to achieve efficient and accurate personalized customization. In addition, in the production link, there is also a lack of production instruction coding and quality traceability mechanism that matches the personalized nutrition plan, and it is impossible to ensure the quality and safety of products. Summary of the Invention

[0006] The purpose of the present invention is to provide a personalized adult nutritional health product customization system and method to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above purpose, the present invention provides the following technical solution: A personalized adult nutritional health product customization method, the method includes:

[0008] F1. Collect the personalized data of users, including biometric data, eating habit data, health goal data, and gene detection data, label the data, and clarify the nutritional requirement labels corresponding to different data categories; perform data preprocessing, including outlier removal and normalization processing, and construct the preprocessed data into a nutritional requirement database;

[0009] F2. Extract single-user data from the nutritional requirement database, reduce the dimension through the principal component analysis method to generate a user feature vector; adopt a sliding time window operation to merge the user feature vectors in consecutive time periods into a dynamic feature matrix;

[0010] F3. Perform cluster analysis on the user group based on the dynamic feature matrix, define a similarity threshold, remove the user data with feature similarity exceeding the threshold, and retain the differentiated user data as a representative sample set; associate and map the representative sample set with a preset nutritional rule library to construct a multi-dimensional nutritional requirement tensor;

[0011] F4. Real-time collect the dynamic feature matrix of the target user, calculate its matching degree with each representative sample in the multi-dimensional nutritional requirement tensor to generate a matching degree vector; use a backpropagation neural network to optimize the matching weight to obtain a weighted matching degree vector;

[0012] F5. Perform a tensor splicing operation on the weighted matching degree vector and the multi-dimensional nutritional requirement tensor to generate a personalized nutrition formula base; iteratively optimize the base through a genetic algorithm to screen out the optimal nutritional combination that meets the constraint conditions;

[0013] F6. According to the user's real-time updated health data, adopt the Kalman filter algorithm to dynamically adjust the component ratio of the nutrition formula base to generate an adaptive nutrition plan;

[0014] F7. Define the nutrition index deviation calculation function to extract the deviation value between the user's current nutrition intake data and the adaptive nutrition plan. If the deviation value exceeds the dynamic threshold, the recipe reconstruction process is triggered;

[0015] F8. Perform knowledge graph association analysis on the reconstructed nutrition plan and the historical plan, generate a user nutrition preference rule set, and update it to the nutrition rule base;

[0016] F9. Output the final nutritional formula and encode it into executable supplement production instructions.

[0017] Preferably, step F1 specifically includes:

[0018] F101. Obtaining raw data through wearable devices, user input and third-party databases, the biometric data includes age, body fat percentage, metabolic rate, the dietary habit data includes dietary records and nutritional intake, and the health goal data includes fat loss, muscle gain or disease prevention goals;

[0019] F102. Label the original data with nutritional requirements, including "high protein requirement", "low sugar requirement", "vitamin deficiency" and "mineral supplementation";

[0020] F103. The interquartile range method was used to eliminate abnormal data, and the data were mapped to the [0,1] interval through minimum-maximum normalization to construct a structured nutritional requirement database.

[0021] Preferably, step F3 specifically includes:

[0022] F301. Define the cosine similarity operator between user dynamic feature matrices, and the calculation formula is:

[0023]

[0024] Among them, X a , X b is the dynamic feature matrix of two users, n is the total dimension of the feature vector, k is the index number of the feature dimension, is the numerical vector of the kth feature dimension in two users,

[0025] F302. Traverse the user group, if the similarity between two users is greater than the threshold γ, randomly remove one of them;

[0026] F303. Classify the retained user data according to the nutritional requirement labels, and perform tensor fusion with the ingredient taboo table and efficacy association table in the nutritional rule library to generate a multidimensional nutritional requirement tensor with a shape of m×p×q, where m is the number of users, p is the nutritional ingredient dimension, and q is the health goal dimension.

[0027] Preferably, the optimization process of the matching weight in step F4 includes:

[0028] F401. Construct a three-layer neural network model, with the input layer being the matching degree vector, the hidden layer using the ReLU activation function, and the output layer being the weight vector;

[0029] F402. Use the historical effect score of the nutrition combination as the supervision signal, and update the network parameters through the gradient descent algorithm to make the weight vector approach the optimal solution.

[0030] Preferably, the constraint conditions of the genetic algorithm in step F5 include:

[0031] F501. The total amount of ingredients does not exceed the daily recommended intake;

[0032] F502. Avoid pharmacological conflicts between ingredients, and generate a penalty function according to the contraindication list;

[0033] F503. Retain the top 10% of individuals with the highest fitness during the iteration process, and generate a new population through crossover and mutation.

[0034] Preferably, step F6 specifically includes:

[0035] F601. Establish a user health state transition equation, which includes variables such as metabolic rate change, exercise amount, and sleep quality;

[0036] F602. Use the adaptive nutrition plan as the observed value, and dynamically adjust the ingredient ratio through the Kalman gain to minimize the error between the predicted value and the actual demand.

[0037] Preferably, the method for setting the dynamic threshold in step F7 is:

[0038] F701. Generate a dynamic threshold interval according to the standard deviation and mean of the user's historical deviation data according to the 3σ principle;

[0039] F702. If the deviation value exceeds the threshold three times in a row, increase the threshold strictness and trigger an emergency reconstruction.

[0040] Preferably, the construction of the knowledge graph in step F8 includes:

[0041] F801. Extract the ingredient, effect, and timestamp data of the user's historical plan to construct a triple relationship;

[0042] F802. Map the triple to a low-dimensional vector through a graph embedding algorithm to generate a user preference subgraph;

[0043] F803. Perform subgraph isomorphism matching between the subgraph and the global nutrition knowledge graph to extract new rules.

[0044] Preferably, the encoding rule of the production instruction in step F9 is as follows:

[0045] F901. Convert the ingredient ratio into a discrete instruction set, including capsule filling amount, powder ratio, and encapsulation parameters;

[0046] F902. Use the hash algorithm to uniquely identify the instruction set and bind it to the user ID for storage in the blockchain.

[0047] Preferably, the present invention further includes a personalized adult nutritional health product customization system, and the system includes:

[0048] Data acquisition and processing module: used to collect personalized data of users, the personalized data includes biometric data, eating habit data, health goal data, and genetic testing data, label the data, clarify the nutritional requirement labels corresponding to different data categories, perform data preprocessing operations such as outlier removal and normalization, and construct the preprocessed data into a nutritional requirement database;

[0049] Feature extraction and dynamic matrix generation module: Extract single-user data from the nutritional requirement database, reduce the dimension through the principal component analysis method to generate a user feature vector, and use the sliding time window operation to merge the user feature vectors in consecutive time periods into a dynamic feature matrix;

[0050] Clustering analysis and tensor construction module: Perform clustering analysis on the user group based on the dynamic feature matrix, define a similarity threshold, remove user data with feature similarity exceeding the threshold, retain the differentiated user data as a representative sample set, and perform an association mapping between the representative sample set and the preset nutritional rule library to construct a multi-dimensional nutritional requirement tensor;

[0051] Matching degree calculation and optimization module: Real-time collect the dynamic feature matrix of the target user, calculate its matching degree with each representative sample in the multi-dimensional nutritional requirement tensor to generate a matching degree vector, and use the backpropagation neural network to optimize the matching weight to obtain a weighted matching degree vector;

[0052] Nutritional formula generation module: Perform a tensor splicing operation on the weighted matching degree vector and the multi-dimensional nutritional requirement tensor to generate a personalized nutritional formula base, and iteratively optimize the base through the genetic algorithm to screen out the optimal nutritional combination that meets the constraint conditions;

[0053] Dynamic adjustment module: According to the user's real-time updated health data, use the Kalman filter algorithm to dynamically adjust the ingredient ratio of the nutritional formula base to generate an adaptive nutritional plan;

[0054] Deviation monitoring and reconstruction module: Define a nutritional index deviation calculation function, extract the deviation value between the user's current nutritional intake data and the adaptive nutritional plan, and if the deviation value exceeds the dynamic threshold, trigger the formula reconstruction process;

[0055] Knowledge graph association and update module: Conduct knowledge graph association analysis on the reconstructed nutrition plan and historical plans, generate a user nutrition preference rule set, and update it to the nutrition rule library;

[0056] Instruction output module: Output the final nutrition formula and encode it into executable supplement production instructions.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] The present invention comprehensively and meticulously depicts the individual characteristics of users by collecting users' biometric data, eating habit data, health goal data, and genetic testing data. For example, by combining the relationship between specific gene loci and nutrient metabolism in genetic testing data, the special needs of users for certain nutrients can be accurately determined. Then, according to biometric data such as body fat percentage and metabolic rate, exclusive nutrition plans are customized for users with different physical conditions. This multi-dimensional data fusion method makes the customization of nutrition plans more in line with the actual needs of users, and the accuracy is greatly improved compared with traditional general nutrition health products.

[0059] Using the Kalman filter algorithm, dynamically adjust the component ratio of the nutrition formula base according to the health data updated in real time by the user. Taking sports enthusiasts as an example, when their exercise intensity and frequency change, the body's requirements for protein, carbohydrates, and electrolytes also change. The present invention can timely capture these changes and automatically adjust the nutrition plan to ensure that users can obtain appropriate nutritional supplements in different health states and continuously maintain a good health state.

[0060] When generating the nutrition formula, the genetic algorithm strictly follows the constraint conditions that the total amount of components does not exceed the daily recommended intake and avoids pharmacological conflicts between components. For example, when designing a nutrition formula containing multiple vitamins and minerals, the algorithm will avoid adding component combinations that may cause adverse reactions according to the taboo list in the nutrition rule library, such as excessive calcium and iron. This not only ensures the health and safety of users, but also selects the optimal nutrition combination through iterative optimization, improving the efficacy of nutrition health products.

[0061] By conducting knowledge graph association analysis on the reconstructed nutrition plan and historical plans, generate a user nutrition preference rule set and update it to the nutrition rule library. As the user data continues to accumulate, the nutrition rule library will become more and more perfect. For example, it is found that users with specific gene types and eating habits have special preferences and good effects on a certain nutrient component combination. Subsequently, when customizing plans for similar users, these new rules can be referred to continuously improve the quality and pertinence of the nutrition plan.

[0062] Encode the final nutritional formula into executable supplement production instructions, and use a hash algorithm to uniquely identify the instruction set and bind it to the user ID for storage in the blockchain. This process realizes the seamless connection from nutritional plan customization to product production, ensuring the accuracy and standardization of production. At the same time, the immutable feature of the blockchain makes the production process traceable, enabling consumers to clearly understand the production information of the product and enhancing their confidence in product quality and safety. Precise personalized customization avoids the situation of excessive or insufficient nutritional components that may occur in traditional general nutritional health products, reducing resource waste. For example, it avoids the large-scale production of nutritional products that are not suitable for specific populations, enabling production resources to be more effectively allocated to user groups with real needs and improving the resource utilization efficiency of the entire industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a working principle diagram of the personalized adult nutritional health product customization method described in the present invention;

[0064] Figure 2 It is a working flow chart of clustering analysis and tensor construction;

[0065] Figure 3 It is a working flow chart of matching weight optimization;

[0066] Figure 4 It is a working flow chart of genetic algorithm constraint condition application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0067] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0068] Please refer to Figures 1-4 , the present invention provides a technical solution: a personalized adult nutritional health product customization method, the method includes:

[0069] F1. Data collection and preprocessing: Collect personalized data of users, covering biometric data, eating habit data, health goal data, and genetic testing data. Label these data to clarify the nutritional requirement tags corresponding to different data categories. Then perform data preprocessing, including outlier removal and normalization processing, and construct the preprocessed data into a nutritional requirement database.

[0070] F2. Feature Extraction and Dynamic Matrix Generation: Extract single-user data from the nutritional requirement database, reduce the dimension using the principal component analysis method, and generate user feature vectors. Adopt a sliding time window operation to merge the user feature vectors in consecutive time periods into a dynamic feature matrix.

[0071] F3. Cluster Analysis and Tensor Construction: Conduct cluster analysis on the user group based on the dynamic feature matrix, define a similarity threshold, eliminate user data with feature similarity exceeding the threshold, and retain the differentiated user data as a representative sample set. Associate and map the representative sample set with the preset nutritional rule library to construct a multi-dimensional nutritional requirement tensor.

[0072] F4. Matching Degree Calculation and Optimization: Collect the dynamic feature matrix of the target user in real time, calculate its matching degree with each representative sample in the multi-dimensional nutritional requirement tensor, and generate a matching degree vector. Use a backpropagation neural network to optimize the matching weights to obtain a weighted matching degree vector.

[0073] F5. Nutritional Formula Generation: Perform a tensor splicing operation on the weighted matching degree vector and the multi-dimensional nutritional requirement tensor to generate a personalized nutritional formula base. Iteratively optimize the base through a genetic algorithm to screen out the optimal nutritional combination that meets the constraint conditions.

[0074] F6. Dynamic Adjustment: According to the user's real-time updated health data, use the Kalman filter algorithm to dynamically adjust the component ratio of the nutritional formula base to generate an adaptive nutritional plan.

[0075] F7. Deviation Monitoring and Reconstruction: Define a nutritional index deviation calculation function, extract the deviation value between the user's current nutritional intake data and the adaptive nutritional plan. If the deviation value exceeds the dynamic threshold, trigger the formula reconstruction process.

[0076] F8. Knowledge Graph Association and Update: Conduct knowledge graph association analysis on the reconstructed nutritional plan and the historical plan to generate a user nutritional preference rule set and update it to the nutritional rule library.

[0077] F9. Instruction Output: Output the final nutritional formula and encode it into executable supplement production instructions.

[0078] The present invention will be further described below in conjunction with Embodiments 1 to 5:

[0079] Embodiment 1:

[0080] During the data collection phase, raw data is obtained through a variety of channels. With the help of wearable devices, users' biometric data, such as age, body fat percentage, metabolic rate and other information, can be collected in real time. The sensors built into wearable devices can accurately measure body fat percentage, and estimate metabolic rate by monitoring the user's heart rate, number of steps and other data combined with specific algorithms. At the same time, a user input interface is provided to facilitate users to enter their own eating habits data, including dietary records, that is, detailed information such as the type and quantity of food consumed by users every day, as well as nutritional intake, such as specific intake of protein, fat, carbohydrates, etc. In addition, data is also obtained from third-party databases to supplement and improve user information, such as obtaining genetic test data, which may contain genetic information related to nutritional metabolism, providing an important basis for subsequent precise customization of nutritional plans.

[0081] The collected raw data is labeled with nutritional requirements. Taking the "high protein requirement" label as an example, if the user's body fat percentage is high and the health goal is to gain muscle, combined with the low protein intake in their dietary records, it can be labeled as "high protein requirement". If the user's blood sugar is high and the sugar intake in the dietary records is high, and the health goal is to prevent diabetes, then it is labeled "low sugar requirement". When the analysis of the user's nutritional intake finds that the long-term intake of a certain vitamin is far below the recommended value, the "vitamin deficiency" label is marked; similarly, when the mineral intake is insufficient, the "mineral supplement" label is marked.

[0082] In the data preprocessing stage, the interquartile range method is used to eliminate abnormal data. First, the quartiles of the data are calculated and the interquartile range is determined. For a certain feature data in the data set, it is sorted from small to large. The lower quartile (Q1) is the value of the 25% position of the sorted data, and the upper quartile (Q3) is the value of the 75% position of the sorted data. The interquartile range (IQR) is the difference between Q3 and Q1. If the data point is less than Q1-1.5*IQR or greater than Q3+1.5*IQR, it is determined to be an outlier and is eliminated. After that, the data is mapped to the [0,1] interval through minimum-maximum normalization. Assuming that the minimum value of a certain feature data is min and the maximum value is max, for any data x in the data set, the result x after normalization is norm The calculation formula is: After such processing, all data are standardized, the influence of data dimension is eliminated, and a structured nutritional requirement database is constructed, laying a good foundation for subsequent analysis and processing.

[0083] Embodiment 2:

[0084] This embodiment focuses on the steps of clustering analysis and tensor construction. Its function is to elaborate on how to perform clustering analysis on user groups, eliminate similar data to retain differentiated samples, and associate the samples with the nutrition rule base to construct a multi-dimensional nutrition requirement tensor, which helps improve the pertinence and accuracy of nutrition plan customization.

[0085] During clustering analysis, first define the cosine similarity operator between user dynamic feature matrices. The calculation formula is:

[0086]

[0087] where, X a 、X b are the dynamic feature matrices of two users, n is the total dimension of the feature vectors, k is the index number of the feature dimension, is the numerical vector of the k-th feature dimension in the two users. Through this formula, the similarity between any two user dynamic feature matrices can be calculated.

[0088] Traverse the user group and set the similarity threshold as γ. If the similarity between two users is greater than the threshold γ, randomly eliminate one of them. For example, in a group of 100 users, after calculation, it is found that the similarity between user A and user B is greater than the threshold. At this time, the random number generator is used to decide whether to eliminate user A or user B, so as to retain the user data with differentiated features.

[0089] Classify the retained user data according to nutrition requirement labels and perform tensor fusion with the ingredient taboo table and efficacy association table in the nutrition rule base. Assume that the ingredient taboo table in the nutrition rule base records the mutual taboo relationships between different nutritional ingredients. For example, taking certain nutritional ingredients simultaneously may cause adverse reactions; the efficacy association table associates the efficacy relationships between nutritional ingredients and different health goals. Taking a simple example, for the user group with "high protein requirement" and "muscle gain" health goals, fuse their data with the relevant information in the nutrition rule base. Finally, generate a multi-dimensional nutrition requirement tensor with the shape of m×p×q, where m is the number of users, p is the nutritional ingredient dimension, and q is the health goal dimension. The multi-dimensional nutrition requirement tensor constructed in this way can comprehensively reflect the requirements of different users in different nutritional ingredients and health goals, providing rich data support for subsequent nutrition plan customization.

[0090] Example 3:

[0091] This embodiment details the process of matching weight optimization. Its function is to optimize by constructing a neural network model and using historical effect scores, making the matching weights more accurate, thereby improving the quality and effectiveness of personalized nutrition formulas.

[0092] First, a three-layer neural network model is constructed. The input layer is a matching degree vector, which is obtained by calculating the matching degree between the dynamic feature matrix of the target user and each representative sample in the multi-dimensional nutritional requirement tensor. The hidden layer uses the ReLU activation function, and the expression of the ReLU function is f(x) = max(0, x). It can effectively solve the problem of gradient disappearance and improve the training efficiency of the neural network. The output layer is a weight vector, which will be used to weight the matching degree vector to obtain a more accurate weighted matching degree vector.

[0093] Using the historical effect scores of nutritional combinations as the supervision signal, the network parameters are updated through the gradient descent algorithm. Assume that the loss function of the network is L, which measures the difference between the current output weight vector of the network and the expected optimal weight vector. The core idea of the gradient descent algorithm is to continuously adjust the network parameters along the negative gradient direction of the loss function, so that the loss function gradually decreases, and thus the weight vector approaches the optimal solution. In the actual training process, the effect scores corresponding to each nutritional combination are obtained from the historical data of nutritional combinations. These scores can be quantitative data such as the improvement of the user's health indicators, such as the reduction of body fat percentage and the increase of muscle mass. These scores are used as the supervision signal and input into the neural network. Through continuous iterative training, the weight and bias parameters of the network are adjusted. Finally, a weight vector that can accurately reflect the importance of each matching degree is obtained, providing strong support for generating a more reasonable personalized nutrition formula.

[0094] Example 4:

[0095] When using the genetic algorithm to iteratively optimize the personalized nutrition formula base, the constraint condition that the total amount of ingredients does not exceed the daily recommended intake must be satisfied first. For example, for the intake of vitamin C, according to the authoritative nutritional guidelines, the daily recommended intake for adults is about 100 milligrams. When generating the nutrition formula, by setting the corresponding upper limit, it is ensured that the content of vitamin C in the formula does not exceed this recommended value, avoiding adverse effects on the human body caused by excessive intake.

[0096] To avoid pharmacological conflicts between ingredients, a penalty function is generated according to the taboo list. Assume that the taboo list in the nutritional rule base records the information that calcium and iron cannot be taken in large amounts at the same time because calcium and iron may compete with each other during the absorption process in the human body. In the genetic algorithm, if the contents of calcium and iron in a certain nutritional formula individual are both relatively high, approaching or exceeding the safe intake range, then the fitness of this individual is reduced according to the penalty function. The penalty function can be designed according to the specific taboo situation. For example, the penalty value is calculated according to the degree to which calcium and iron exceed the safe range, so that the probability of nutritional formula individuals containing potential pharmacological conflicts being eliminated in the selection process of the genetic algorithm increases, thereby guiding the algorithm to generate a safer and more reasonable nutrition formula.

[0097] During the iteration process, the top 10% of individuals with the highest fitness are retained, and a new population is generated through crossover and mutation. Fitness is an indicator to measure the quality of each nutritional formula individual, and it can be calculated based on factors such as the degree to which the nutritional formula meets the user's health goals and the rationality of the ingredients. In each generation of iteration, the top 10% of individuals with the highest fitness are selected and directly retained for the next generation to ensure that excellent nutritional formula characteristics are not lost. At the same time, through the crossover operation, part of the genes of two parent individuals are exchanged to generate new offspring individuals; through the mutation operation, certain genes of individuals are randomly changed with a certain probability to introduce new gene combinations, increasing the diversity of the population, enabling the genetic algorithm to search in a wider solution space, and improving the possibility of finding the optimal nutritional combination.

[0098] Example 5:

[0099] This example covers the relevant content regarding dynamic adjustment, deviation monitoring and reconstruction, knowledge graph association and update, and instruction output. Its role is to enable the nutritional plan to be dynamically adjusted according to the user's real-time health data, continuously optimize the nutritional plan, and feedback the optimization experience to the nutritional rule base. At the same time, executable production instructions are generated to realize the complete process from plan customization to product production.

[0100] In terms of dynamic adjustment, a user health state transition equation is established, which includes variables such as metabolic rate change, exercise amount, and sleep quality. For example, the change in metabolic rate can be estimated through data such as the user's weight change and exercise consumption over a period of time; the exercise amount can obtain information such as the user's exercise steps, exercise type, and exercise duration from wearable devices; the sleep quality can be evaluated by monitoring data such as the user's sleep duration and sleep stages. Taking the adaptive nutritional plan as the observed value, the component ratio is dynamically adjusted through the Kalman gain to minimize the error between the predicted value and the actual demand. The Kalman filter algorithm is a commonly used optimal estimation method, which can continuously update the estimation of the system state according to the state equation and observation equation of the system, combined with the estimation value of the previous moment and the observation value of the current moment. In the adjustment of the nutritional plan, the Kalman filter algorithm can be used to dynamically adjust the ratio of various nutritional components in the nutritional formula base according to the user's real-time health data to better meet the user's nutritional needs.

[0101] In the deviation monitoring and reconstruction phase, a function for calculating the deviation of nutritional indicators is defined to extract the deviation value between the user's current nutritional intake data and the adaptive nutritional plan. For example, calculate the difference between the actual intake of nutrients such as protein and vitamins by the user and the recommended intake in the adaptive nutritional plan. According to the standard deviation and mean of the user's historical deviation data, a dynamic threshold range is generated according to the 3σ principle. If the deviation value exceeds the threshold three times in a row, the threshold strictness is increased and an emergency reconstruction is triggered. Suppose through the analysis of the user's historical data, the mean of the intake deviation of a certain nutrient is μ and the standard deviation is σ, then the dynamic threshold range is [μ - 3σ, μ + 3σ]. When the user's nutritional intake deviation exceeds this range three times in a row, the threshold range is appropriately narrowed, the threshold strictness is increased, and the formula reconstruction process is started to ensure that the nutritional plan always meets the user's actual needs.

[0102] Regarding the association and update of the knowledge graph, extract the component, effect, and timestamp data of the user's historical plan to construct a triple relationship. For example, a triple can be (vitamin C, enhance immunity, 2023-01-01), indicating that on January 1, 2023, after the user ingested vitamin C, it had the effect of enhancing immunity. The triple is mapped to a low-dimensional vector through a graph embedding algorithm to generate a user preference subgraph. The graph embedding algorithm can transform complex graph structure data into low-dimensional vector representations, facilitating computer processing and analysis. The subgraph is subjected to subgraph isomorphism matching with the global nutritional knowledge graph to extract new rules. For example, if a new relationship between a certain combination of nutrients and a specific health effect is found in the user preference subgraph and this relationship is not recorded in the global nutritional knowledge graph, then it is extracted as a new rule and updated to the nutritional rule library to provide more knowledge support for subsequent nutritional plan customization.

[0103] In the instruction output stage, the component ratios are converted into a discrete instruction set, including capsule filling amounts, powder ratios, and encapsulation parameters. For example, for a health product containing multiple nutrients, if the ratio of a certain component in the nutritional formula is 10%, according to the production process and product specifications, it is converted into a specific capsule filling amount, such as filling 0.1 grams of this component in each capsule. The hash algorithm is used to uniquely identify the instruction set and bind it to the user ID for storage in the blockchain. The hash algorithm can convert the instruction set into a hash value of a fixed length, ensuring the uniqueness and immutability of the instruction set. Binding the hash value to the user ID and storing it in the blockchain, using the decentralized and immutable characteristics of the blockchain, ensures the security and traceability of production instructions, guaranteeing the quality and accuracy of personalized nutritional health products from the production link.

[0104] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, such that a process, method, article or apparatus comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or apparatus.

[0105] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for customizing personalized adult nutritional and health products, characterized in that, The steps are as follows: F1. Collect the personalized data of users, including biometric data, eating habit data, health goal data and genetic test data, label the data, and clarify the nutrition requirement labels corresponding to different data categories; perform data preprocessing, including outlier removal and normalization, and construct a nutrition requirement database with the preprocessed data; F2. Extract single-user data from the nutrition requirement database, reduce the dimension through the principal component analysis method, and generate user feature vectors; Adopt a sliding time window operation to merge the user feature vectors in consecutive time periods into a dynamic feature matrix; F3. Conduct a clustering analysis on the user group based on the dynamic feature matrix, define a similarity threshold, remove the user data with feature similarity exceeding the threshold, and retain the differentiated user data as a representative sample set; perform an association mapping between the representative sample set and a preset nutrition rule library to construct a multi-dimensional nutrition requirement tensor; F4. Collect the dynamic feature matrix of the target user in real time, calculate its matching degree with each representative sample in the multi-dimensional nutrition requirement tensor, and generate a matching degree vector; Optimize the matching weights using a backpropagation neural network to obtain a weighted matching degree vector; F5. Perform a tensor splicing operation on the weighted matching degree vector and the multi-dimensional nutrition requirement tensor to generate a personalized nutrition formula base; iteratively optimize the base through a genetic algorithm to screen out the optimal nutrition combination that meets the constraint conditions; F6. According to the health data updated by the user in real time, adopt the Kalman filter algorithm to dynamically adjust the component ratio of the nutrition formula base to generate an adaptive nutrition plan; F7. Define a nutrition index deviation calculation function, extract the deviation value between the user's current nutrition intake data and the adaptive nutrition plan, and if the deviation value exceeds the dynamic threshold, trigger the formula reconstruction process; F8. Conduct a knowledge graph association analysis on the reconstructed nutrition plan and the historical plan to generate a user nutrition preference rule set, and update it to the nutrition rule library; F9. Output the final nutrition formula and encode it into an executable supplement production instruction.

2. The method according to claim 1, wherein Step F1 specifically includes: F101. Obtain the original data through wearable devices, user input and third-party databases. The biometric data includes age, body fat percentage, and metabolic rate. The eating habit data includes dietary records and nutrition intake. The health goal data includes fat loss, muscle gain or disease prevention goals; F102. Label the original data with nutrition requirement labels, including "high protein requirement", "low sugar requirement", "vitamin deficiency", and "mineral supplement"; F103. Remove abnormal data using the interquartile range method and map the data to the [0,1] interval through min-max normalization to construct a structured nutrition requirement database.

3. The method according to claim 1, wherein Step F3 specifically includes: F301. Define a cosine similarity operator between user dynamic feature matrices, and the calculation formula is: Among them, X a , X b are the dynamic feature matrices of two users, n is the total dimension of the feature vectors, k is the index number of the feature dimension, is the numerical vector of the k-th feature dimension in the two users, F302. Traverse the user group. If the similarity between two users is greater than the threshold γ, randomly remove one of them; Classify the retained user data according to the nutritional requirement labels, and perform tensor fusion with the ingredient taboo table and the efficacy association table in the nutritional rule base to generate a multi-dimensional nutritional requirement tensor with the shape of m×p×q, where m is the number of users, p is the dimension of nutritional ingredients, and q is the dimension of health goals.

4. The method according to claim 1, wherein The optimization process of the matching weight in step F4 includes: F401. Construct a three-layer neural network model, where the input layer is the matching degree vector, the hidden layer uses the ReLU activation function, and the output layer is the weight vector; F402. Use the historical effect score of the nutritional combination as the supervision signal, and update the network parameters through the gradient descent algorithm to make the weight vector approach the optimal solution.

5. The method according to claim 1, characterized in that The constraint conditions of the genetic algorithm in step F5 include: F501. The total amount of ingredients does not exceed the daily recommended intake; F502. Avoid pharmacological conflicts between ingredients, and generate a penalty function according to the taboo table; F503. Retain the top 10% of individuals with the highest fitness during the iteration process, and generate a new population through crossover and mutation.

6. The method according to claim 1, characterized in that Step F6 specifically includes: F601. Establish a user health state transition equation, which includes variables such as metabolic rate change, exercise amount, and sleep quality; F602. Use the adaptive nutritional plan as the observed value, and dynamically adjust the ingredient ratio through the Kalman gain to minimize the error between the predicted value and the actual demand.

7. The method according to claim 1, wherein The method for setting the dynamic threshold in step F7 is: F701. Generate a dynamic threshold interval according to the standard deviation and mean of the user's historical deviation data according to the 3σ principle; F702. If the deviation value exceeds the threshold three times in a row, increase the threshold strictness and trigger an emergency reconstruction.

8. The method according to claim 1, characterized in that The construction of the knowledge graph in step F8 includes: F801. Extract the ingredient, effect, and timestamp data of the user's historical plan, and construct a triple relationship; F802. Map the triple to a low-dimensional vector through a graph embedding algorithm to generate a user preference subgraph; F803. Perform subgraph isomorphism matching between the subgraph and the global nutritional knowledge graph to extract new rules.

9. The method according to claim 1, wherein The encoding rule of the production instruction in step F9 is: F901. Convert the ingredient ratio into a discrete instruction set, including capsule filling amount, powder ratio, and encapsulation parameters; F902. Use the hash algorithm to uniquely identify the instruction set and bind it to the user ID for storage in the blockchain.

10. A personalized adult nutritional health product customization system, characterized in that, Including: Data acquisition and processing module: used to collect the user's personalized data, the personalized data includes biometric data, eating habit data, health goal data, and gene detection data, label the data, clarify the nutritional requirement labels corresponding to different data categories, perform data preprocessing operations such as outlier removal and normalization, and construct the preprocessed data into a nutritional requirement database; Feature extraction and dynamic matrix generation module: extract single-user data from the nutritional requirement database, reduce the dimension through the principal component analysis method to generate a user feature vector, and use the sliding time window operation to merge the user feature vectors in consecutive time periods into a dynamic feature matrix; Clustering Analysis and Tensor Construction Module: Conduct clustering analysis on user groups based on the dynamic feature matrix, define a similarity threshold, eliminate user data with feature similarity exceeding the threshold, retain the differentiated user data as a representative sample set, perform an association mapping between the representative sample set and the preset nutrition rule base, and construct a multi-dimensional nutrition requirement tensor; Matching Degree Calculation and Optimization Module: Collect the dynamic feature matrix of the target user in real time, calculate its matching degree with each representative sample in the multi-dimensional nutrition requirement tensor to generate a matching degree vector, and use a backpropagation neural network to optimize the matching weights to obtain a weighted matching degree vector; Nutritional Formula Generation Module: Perform a tensor splicing operation on the weighted matching degree vector and the multi-dimensional nutrition requirement tensor to generate a personalized nutritional formula base, and iteratively optimize the base through a genetic algorithm to screen out the optimal nutritional combination that meets the constraint conditions; Dynamic Adjustment Module: According to the user's real-time updated health data, use the Kalman filter algorithm to dynamically adjust the composition ratio of the nutritional formula base to generate an adaptive nutrition plan; Deviation Monitoring and Reconstruction Module: Define a nutritional index deviation calculation function, extract the deviation value between the user's current nutritional intake data and the adaptive nutrition plan, and trigger the formula reconstruction process if the deviation value exceeds the dynamic threshold; Knowledge Graph Association and Update Module: Conduct knowledge graph association analysis on the reconstructed nutrition plan and the historical plan to generate a user nutrition preference rule set and update it to the nutrition rule base; Instruction Output Module: Output the final nutritional formula and encode it into an executable supplement production instruction.

Images