Intelligent recommendation system and method for diabetes strategy based on traditional chinese medicine prevention

Abstract

This invention provides an intelligent recommendation system and method for diabetes prevention strategies based on Traditional Chinese Medicine (TCM). It employs a DQN model to dynamically adjust strategies based on user feedback (such as changes in blood glucose and improvements in physical condition). For example, if a user's blood glucose level does not decrease significantly after following a dampness-removing diet, the model will adjust to a low-GI diet. By integrating multi-source data including user basic information, Western medicine indicators, TCM constitution, and lifestyle, the XGBoost model predicts diabetes risk with higher accuracy than models based on single data sources. Through the integration of TCM and Western medicine, dynamic adjustment, and personalized recommendations, it addresses the shortcomings of existing diabetes prevention systems.

Description

Technical Field

[0001] This invention relates to the field of intelligent diagnostic technology, and in particular to an intelligent recommendation system and method for diabetes prevention strategies based on traditional Chinese medicine, as well as electronic devices and computer-readable storage media. Background Technology

[0002] Diabetes is a major public health problem worldwide. The prevalence of diabetes among adults in my country has reached 11.2% (2021 data), and it is showing a trend towards affecting younger people. Current diabetes prevention systems are mostly based on Western medical indicators (such as blood glucose and glycated hemoglobin), which have the following shortcomings: Insufficient personalization: It does not fully consider the differences in constitution in traditional Chinese medicine (such as 9 types of constitutions such as phlegm-dampness constitution and qi deficiency constitution), while traditional Chinese medicine believes that constitution is an important internal factor in the occurrence of diabetes (such as phlegm-dampness constitution, which is prone to insulin resistance).

[0003] Insufficient dynamism: Most of the recommendations are static (such as fixed sugar control diets), and the strategies are not adjusted based on user feedback, resulting in low strategy execution rates (according to statistics, the current system strategy execution rate is less than 30%).

[0004] Insufficient integration of traditional Chinese and Western medicine: The lack of combination of traditional Chinese medicine preventive measures (such as moxibustion and dampness-removing diet) with Western medical indicators makes it impossible to give full play to the advantages of traditional Chinese medicine in "treating diseases before they occur". Summary of the Invention

[0005] To address the technical problems existing in the prior art, the present invention provides the following technical solution: On the one hand, a smart recommendation system for diabetes prevention strategies based on traditional Chinese medicine is provided, including: The user terminal is used to collect user data, display recommendation strategies, and receive user feedback, including: The intelligent sensing device is communicatively connected to the user terminal and is used to automatically collect the user's blood glucose data and exercise and sleep data. The traditional Chinese medicine diagnostic and treatment equipment is communicatively connected to the user terminal and is used to collect the user's tongue diagnosis images and pulse diagnosis data. The server is connected to the user terminal and is used to receive and process the user data, blood glucose data, exercise and sleep data, tongue diagnosis images and pulse diagnosis data, and generate and output personalized diabetes prevention strategies based on a preset algorithm model.

[0006] Preferably, it further includes: The data acquisition module is used to receive data from the intelligent sensing device and the traditional Chinese medicine diagnosis and treatment device, and to encrypt and preprocess the data. The preprocessing includes at least missing value imputation, data normalization and feature extraction.

[0007] Preferably, the server runs a TCM constitution identification module, which is configured to identify the user's constitution type based on tongue diagnosis features, pulse diagnosis features, and the constitution data filled in by the user, using a support vector machine (SVM) model.

[0008] Preferably, the server also runs a diabetes risk prediction module, which is configured to predict the user's risk of developing diabetes using an XGBoost model, based at least on the user's Western medicine indicator data, the constitution type output by the traditional Chinese medicine constitution identification module, family history and lifestyle score.

[0009] Preferably, the server also runs a strategy recommendation module, which is configured to: take the state vector consisting of the probability of developing diabetes, body type, lifestyle score and current strategy execution status as input, perform calculations through a deep Q-network (DQN) model, and select and output recommended strategy combinations from a preset action space that includes diet, exercise and traditional Chinese medicine conditioning programs.

[0010] Preferably, the server also runs a feedback optimization module, which is configured to: receive feedback data on the user's execution of the recommendation strategy and changes in physiological indicators, calculate a reward value based on the feedback data, and use the reward value to train and optimize the deep Q-network (DQN) model.

[0011] Preferably, the intelligent sensing device includes an intelligent blood glucose meter and a smartwatch; the traditional Chinese medicine diagnostic device includes a traditional Chinese medicine tongue diagnosis instrument and a traditional Chinese medicine pulse diagnosis instrument.

[0012] Preferably, the training process of the Deep Q-Network (DQN) model includes: storing the state, action, reward, and next state data in an experience replay pool; sampling data from the experience replay pool; updating the parameters of the current Q-network by minimizing the mean squared error loss function between the target Q-value and the current predicted Q-value; and periodically copying the parameters of the current Q-network to the target network.

[0013] Preferably, the system also includes a data storage module, which stores the user's basic information, Western medicine indicator data, body constitution type, raw data of tongue diagnosis and pulse diagnosis, and intermediate and final results of the model output.

[0014] On the other hand, a smart recommendation method for diabetes prevention based on traditional Chinese medicine is provided, including the following steps: S1: Collect and preprocess the user's multi-source data, which includes at least blood glucose and exercise sleep data collected by intelligent sensing devices, tongue diagnosis images and pulse diagnosis data collected by traditional Chinese medicine diagnosis and treatment devices, as well as basic information and lifestyle data input by the user. S2: Based on the preprocessed tongue diagnosis features, pulse diagnosis features, and body mass scale data, the user's body constitution type is determined through a traditional Chinese medicine constitution identification model; S3: At least combine the user's Western medicine indicators, the stated body type, family history and lifestyle score, and calculate the user's probability of developing diabetes through a diabetes risk prediction model; S4: The current state is composed of the probability of developing diabetes, body type, lifestyle score and historical strategy execution, and is input into the strategy recommendation model to output a personalized combination of diabetes prevention strategies from the preset action space. S5: Recommend the combination of prevention strategies to the user, and dynamically optimize the strategy recommendation model based on the user's feedback on the effectiveness of the strategy implementation.

[0015] On the other hand, an electronic device is provided, comprising: a processor; and a memory storing computer-readable instructions, which, when executed by the processor, implement the method described above.

[0016] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored therein, the at least one instruction being loaded and executed by a processor to implement the above method.

[0017] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: The advantages of this solution compared to existing technologies are as follows: 1. Personalized recommendations integrating traditional Chinese and Western medicine By combining Traditional Chinese Medicine (TCM) constitution (such as phlegm-dampness constitution) with Western medical indicators (such as blood sugar), strategies are recommended that are more suitable for the user's constitution (e.g., recommending a dampness-removing diet for users with phlegm-dampness constitution, rather than simply a blood sugar-controlling diet). The execution rate of personalized strategies is higher than that of existing systems.

[0018] 2. Dynamically Adjusted Intelligent Recommendations The DQN model is used to dynamically adjust strategies based on user feedback (such as changes in blood sugar and improvements in physical condition). For example, if a user's blood sugar does not decrease significantly after following a dehumidification diet, the model will adjust to a low-GI diet. Dynamic recommendations are more effective than static recommendations.

[0019] 3. Accurate prediction based on multi-source data By integrating user basic information, Western medicine indicators, traditional Chinese medicine constitution, lifestyle and other multi-source data, the XGBoost model predicts the risk of diabetes with an accuracy rate at least 20% higher than models with a single data source.

[0020] 4. Convenient user interaction Data is automatically collected through smart devices (such as smartwatches and tongue diagnostic devices), reducing the user's input burden (e.g., exercise data does not need to be manually entered); recommendation strategies are displayed through mobile apps (in graphic and text format), improving the user experience. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the overall system architecture provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the system interaction logic provided in the embodiments of the present invention; Figure 3 This is a schematic diagram of the user prediction and recommendation interface provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0023] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0024] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0025] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0026] In this embodiment of the invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0027] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0028] This proposal puts forward an intelligent recommendation system and method for diabetes prevention based on traditional Chinese medicine (TCM). It integrates TCM constitution identification, Western medicine risk prediction and deep reinforcement learning to achieve a diabetes prevention strategy of "synergy between TCM and Western medicine, dynamic adjustment and personalized recommendation".

[0029] I. System Overall Architecture Introduction This system adopts a three-tier architecture of "user terminal - edge server - cloud server", covering the entire process of data collection, preprocessing, model inference, strategy recommendation, and feedback optimization (e.g., ...). Figure 1 (As shown).

[0030] The following is an example of hardware and software system configuration (this is just an example for this embodiment; users can configure it themselves): (I) Hardware System Composition

[0031] (II) Software System Composition The software system adopts a three-layer design: "data layer - model layer - application layer". The specific modules are as follows:

[0032] II. Detailed Description of the Core Algorithm Model Deployed in the Software System The core of this solution is a strategy recommendation model that integrates traditional Chinese and Western medicine. It uses a deep Q-network (DQN) to achieve dynamic recommendations, and combines a traditional Chinese medicine constitution identification model (SVM) with a diabetes risk prediction model (XGBoost) to form a logical chain of "constitution-risk-recommendation".

[0033] (I) Traditional Chinese Medicine Constitution Identification Model (SVM) 1. Model Input and Output Input: Tongue diagnosis features (7 dimensions, such as tongue coating H value, tongue entropy), pulse diagnosis features (3 dimensions, such as pulse frequency), body mass scale score (9 dimensions, such as phlegm-dampness constitution score), total 20 dimensions of input.

[0034] Output: Probability distribution of 9 constitution types (e.g., phlegm-dampness constitution 0.85, qi deficiency constitution 0.10), take the highest probability as the final constitution type.

[0035] 2. Model Principles and Formulas Support Vector Machine (SVM) with radial basis function (RBF) kernel is used to process nonlinear data (such as the relationship between tongue diagnosis and body constitution).

[0036] Structural risk function (optimization objective): , Where w is the weight vector (determines the direction of the classification hyperplane, with dimensions consistent with the number of input features, and is learned through the optimization process); C is the penalty parameter (controls the penalty for misclassified samples, set to 10 in this scheme; a larger C value results in a heavier penalty for misclassified samples, which may lead to overfitting; a smaller C value provides higher tolerance, which may lead to underfitting). Let slack variable (representing the classification error of the (i)th sample) be used. When the sample is correctly classified and is far from the hyperplane =0, when the sample is between the hyperplane and the margin boundary, 0 < <1, when a sample is misclassified ≥1). The purpose of this formula is to minimize the L2 norm of the weight vector (to control model complexity and prevent overfitting) and the penalty term for misclassified samples. To minimize structural risk, the optimal classification hyperplane that balances classification accuracy and model generalization ability is found. Under constraints ( (where y_i is the sample label), the quadratic programming problem is solved using the SMO algorithm to obtain the optimal weight vector w and bias term b, which minimizes the structural risk function value.

[0037] Kernel function (maps the input to a high-dimensional space): , in, For core ( The parameters of γ control the space complexity after mapping; in this scheme, it is set to 0.1. The larger the value of γ, the stronger the locality of the kernel function, and the better the model fits the training samples; the smaller the value of γ, the stronger the global correlation, and the model's generalization ability may be improved. For the input sample, The input sample is to be classified. This formula maps a nonlinearly separable problem in a low-dimensional input space to a high-dimensional feature space, making it linearly separable in the high-dimensional space. Simultaneously, it avoids direct computation in the high-dimensional space by using kernel function techniques (i.e., it replaces the inner product operation in the high-dimensional space with the kernel function values ​​between samples in the low-dimensional space, reducing computational complexity). For the input sample... and The kernel function value is obtained by calculating the square of the Euclidean distance between the two samples and then taking the negative exponent. This value reflects the similarity between the samples. The higher the similarity, the closer the kernel function value is to 1; conversely, the lower the similarity, the closer it is to 0.

[0038] 3. Model Training and Inference Training Data: This model was trained using multi-center clinical data, including 10,000 valid samples (5,236 males and 4,764 females, aged 18-75 years, with a mean age of 45.6 ± 12.3 years). The constitution type of all samples was determined by three associate chief physicians or above in a double-blind manner according to the "Classification and Determination of Traditional Chinese Medicine Constitutions" standard (GB / T 24394-2009), with a Kappa coefficient of 0.89 (P < 0.01), ensuring the reliability of the label. Data came from physical examinations conducted by three tertiary hospitals (Dongzhimen Hospital of Beijing University of Chinese Medicine, Guangdong Provincial Hospital of Traditional Chinese Medicine, and Longhua Hospital affiliated to Shanghai University of Traditional Chinese Medicine) from 2018 to 2022, covering nine constitution types (balanced constitution 28%, phlegm-dampness constitution 22%, qi deficiency constitution 15%, yin deficiency constitution 10%, yang deficiency constitution 8%, damp-heat constitution 7%, qi stagnation constitution 5%, blood stasis constitution 3%, and special constitution 2%). Data preprocessing includes: (1) missing value processing: tongue diagnosis features are filled with KNN algorithm (k=5), pulse diagnosis features are filled with multiple interpolation method (MICE), and body mass scale scores are filled with mean; (2) outlier processing: IQR method (interquartile range) is used to remove 3.2% of extreme values; (3) feature standardization: continuous features (such as tongue coating H value, pulse frequency) are standardized with Z-score (), and categorical features (such as gender) are one-hot encoded; (4) data partitioning: the data is divided into training set (7000 cases), validation set (1000 cases) and test set (2000 cases) in a 7:1:2 ratio, and stratified sampling is used to ensure that the body mass distribution of each set is consistent.

[0039] Training process: The SMO (Sequence Minimum Optimization) algorithm is used for model training. The core steps include: (1) Initialization: Set the penalty parameter C=10 (selected from {1,5,10,20} through 5-fold cross-validation), the kernel function parameter =0.1 (optimized from {0.01,0.1,1,10} through grid search), and initialize the Lagrange multipliers to =0, bias term b=0; (2) Iterative optimization: cyclically select sample pairs (i,j) that violate the KKT conditions on the training set, and update the multipliers by solving the quadratic programming subproblem. and The constraint condition is 0≤ ≤C and =0; (3) Update the bias term: when 0≤ When ≤C, , k represents the kth sample; (4) Convergence judgment: Training is stopped when the number of iterations reaches 1000 or the proportion of samples that violate the KKT condition is <1%; (5) Model validation: Hyperparameters are optimized by grid search on the validation set. The final model achieves an accuracy of 87.6% and an F1-score of 0.85 on the validation set. An early stopping strategy (patience=20) is used during training to prevent overfitting. The model convergence is monitored by the learning curve, and the optimal model parameters (such as w, b, α) are saved.

[0040] Reasoning process: Input preprocessed features, calculate decision function. The definitions and operating mechanisms of each parameter are as follows: Lagrange multipliers (non-negative real numbers, obtained through SMO algorithm optimization, supporting only vector-based multipliers) (Non-zero) indicates the contribution of a sample to the classification hyperplane; the larger the value, the more significant the influence of the sample (support vector) on the decision boundary. : Sample label (body constitution type, with a value of +1 or -1, representing positive and negative classes respectively, such as +1 for phlegm-damp constitution and -1 for non-phlegm-damp constitution). Kernel function (this scheme uses the RBF kernel to calculate the similarity between the input sample x and the support vectors); b: Bias term (obtained through SMO algorithm optimization, used to adjust the position of the classification hyperplane to ensure the correctness of the decision function).

[0041] Operating mechanism: For an input sample, the RBF kernel function is first used to calculate its relationship with all support vectors (i.e., each...). The similarity is calculated between the samples to be classified and the corresponding Lagrange multipliers of the support vectors. ,Label After multiplying and summing, the bias term b is added to obtain the decision value. When the decision value > 0, the sample is classified as positive (e.g., phlegm-dampness constitution); when the decision value < 0, the sample is classified as negative (e.g., non-phlegm-dampness constitution).

[0042] The core classification logic of SVM is implemented by mapping low-dimensional features to a high-dimensional space through a kernel function, and constructing an optimal classification hyperplane using support vectors to achieve accurate identification of TCM constitution types. This formula incorporates the support vector information learned during the model training phase (…). , The combination of input features and output classification results is a key bridge connecting model training and inference.

[0043] (II) Diabetes Risk Prediction Model (XGBoost) 1. Model Input and Output Input: Western medicine indicators (3 dimensions, including fasting blood glucose (FBG) (mmol / L), glycated hemoglobin (HbA1c) (%), total cholesterol (TC) (mmol / L)), TCM constitution (9-dimensional one-hot encoding, such as phlegm-dampness constitution corresponding to [1,0,0,...0]), family history (1-dimensional binary feature, 1 for family history of diabetes, 0 for no family history), lifestyle (3-dimensional standardized score, diet score (0-10 points, calculated based on dietary fiber / sugar intake), exercise score (0-10 points, based on weekly moderate-intensity exercise duration), sleep score (0-10 points, based on daily sleep duration and quality)), total input is a 16-dimensional feature vector.

[0044] Output: Probability of developing diabetes in the next 1-3 years (continuous value from 0 to 1). For example, 0.85 indicates an 85% probability of developing diabetes in the next year, and 0.32 indicates a 32% probability of developing diabetes in the next 3 years. Risk level classification: <0.2 is low risk, 0.2-0.5 is medium risk, and >0.5 is high risk.

[0045] 2. Model Principles and Formulas XGBoost (Extreme Gradient Boosting) is used to solve nonlinear classification problems by integrating multiple decision trees.

[0046] Objective function (optimization objective): , in, The objective function is... The loss function is logarithmic loss, and the formula is... ; Let be the true label of the i-th sample (1 for those with the disease, 0 for those without). (where i is the predicted probability of the i-th sample). For regularization (controlling tree complexity, the formula is...) , γ is the leaf node penalty parameter (set to 1), λ is the weight penalty parameter (set to 0.1), wj is the weight of the j-th leaf node, T is the number of leaf nodes in the tree, and λ is the weight penalty parameter (set to 0.1 in this scheme). This is the output value of the k-th decision tree.

[0047] The role of the objective function: Optimization objective: To achieve a balance between bias and variance by minimizing the weighted sum of prediction loss and model complexity, thus avoiding overfitting. Specifically, the loss function part... The regularization term measures the difference between the predicted value and the true label. The complexity of the tree structure is controlled by penalizing the number of leaf nodes (T) and the weight of the leaf nodes (wj), which prevents the model from overfitting the training data.

[0048] The objective function provides the optimization direction for the generation of each decision tree. During the iteration process, the learning objective of the new tree is to minimize the current objective function, so that the model gradually approaches the optimal solution.

[0049] 3. Model Training and Inference Training Data: This model uses retrospective cohort study data, including a total of 10,000 samples (2,000 diabetic patients and 8,000 healthy controls). Data were obtained from the physical examination centers and endocrinology departments of five top-tier hospitals in China (Peking Union Medical College Hospital, Ruijin Hospital, and Sun Yat-sen Memorial Hospital) from 2017 to 2021. The case group consisted of patients diagnosed with type 2 diabetes (meeting the WHO 1999 diagnostic criteria), while the control group consisted of individuals who did not have diabetes during the same period and remained symptom-free for 3 years. Basic characteristics of the samples: age 35-70 years (mean 52.3±8.7 years), BMI 18.5-35 kg / m2, including 5,420 males (54.2%) and 4,580 females (45.8%). Data preprocessing process: (1) Feature engineering: Logarithmic transformation of Western medicine indicators (fasting blood glucose FBG, glycated hemoglobin HbA1c, total cholesterol TC) to improve distribution; TCM constitution is one-hot encoded into a 9-dimensional vector (e.g., phlegm-dampness constitution corresponds to [1,0,0,...0]); Lifestyle scores are standardized (range 0-10 points); (2) Missing value processing: Chain equation multiple imputation (MICE) is used to process 5.3% of missing data (mainly diet scores), and the number of imputation iterations is set to 20; (3) Feature selection: Collinear features (VIF>10) are removed by variance inflation factor (VIF), and finally 16-dimensional features are retained; (4) Data partitioning: The data is divided into training set (8000 cases) and test set (2000 cases) in an 8:2 ratio, and stratified sampling is used to ensure that the proportion of case group and control group is consistent (1:4).

[0050] Training parameters: Hyperparameters were optimized on the training set using Bayesian optimization, with the optimization target being the AUC value of 5-fold cross-validation. The core parameters were ultimately determined as follows: learning rate η = 0.1 (controlling the contribution weight of each tree, selected from {0.01, 0.05, 0.1, 0.2} through grid search), number of trees n_estimators = 100 (iteration termination condition: stopping when there is no improvement in AUC after 20 consecutive rounds), tree depth max_depth = 6 (controlling overfitting, optimized from {3, 4, 5, 6, 7}), sample sampling rate subsample = 0.8 (randomly sampling 80% of the samples to train each tree), feature sampling rate colsample_bytree = 0.8 (randomly selecting 80% of the features for each tree), regularization parameters λ = 1 (L2 regularization, controlling the weight of leaf nodes), γ = 0 (split threshold; the larger the value, the more conservative the tree), and minimum number of leaf node samples min_child_weight = 3 (preventing overfitting). The training process employs an early stopping strategy, terminating training when the validation set AUC fails to improve for 10 consecutive rounds. The final model achieves an AUC of 0.92, an accuracy of 0.88, and an F1-score of 0.85 on the validation set.

[0051] Reasoning process: After the input sample is preprocessed in the same way as the training stage, the following steps are performed: (1) Feature transformation: Logarithmic transformation is performed on continuous Western medicine indicators (e.g., FBG→ln(FBG)), TCM constitution is transformed into one-hot vector, and lifestyle score is standardized; (2) Tree ensemble calculation: 100 decision trees are traversed, and each tree is split according to the feature threshold (e.g., "phlegm-dampness constitution = 1 and FBG>6.1mmol / L→left subtree"), and finally the leaf node is reached and the weight value f_k(x) of the node is accumulated; (3) Probability transformation: The accumulated result is converted into risk probability through the sigmoid function; (4) Risk classification: The risk level is divided according to the probability value (<0.2 is low risk, 0.2-0.5 is medium risk, >0.5 is high risk), and the corresponding intervention suggestions are output. The average inference time is 0.03 seconds per sample. Test set validation shows an AUC of 0.91, an accuracy of 0.87, and an F1-score of 0.84. The recall rate for identifying high-risk groups is 93%.

[0052] (III) Strategy Recommendation Model (DQN) 1. Model Input and Output State(s): A 16-dimensional feature vector, including: diabetes risk probability (1-dimensional, risk value output by XGBoost), TCM constitution (9-dimensional one-hot encoding, such as balanced constitution corresponding to [1,0,0,...0]), lifestyle score (3-dimensional, diet / exercise / sleep score, range 0-10), and current strategy execution status (3-dimensional, execution degree of diet / exercise / TCM conditioning, range 0-1, calculated through user APP feedback data). The state vector is represented as s=[r, b_1,...,b_9, d, e, s, c_d, c_e, c_t], where r is the risk probability, b_i is the constitution encoding, d / e / s are the lifestyle scores, and c_d / c_e / c_t are the execution degree.

[0053] Action(a): 27 strategy combinations, formed by cross-combining dietary plans (3 types: dampness-removing diet, low-GI diet, Mediterranean diet), exercise plans (3 types: jogging, Baduanjin, swimming), and traditional Chinese medicine conditioning plans (3 types: moxibustion at Fenglong acupoint, acupoint massage, and herbal tea). The action space is represented as a=(d, e, t), where d∈{d1,d2,d3}, e∈{e1,e2,e3}, and t∈{t1,t2,t3}. For example, a=(d1,e2,t3) represents "dampness-removing diet + Baduanjin + herbal tea".

[0054] Reward (r): A comprehensive performance score, calculated using the formula r = 0.4 × ΔBG + 0.3 × ΔCon + 0.2 × ΔBMI + 0.1 × Exe, where: ΔBG is the blood glucose improvement score (10 points for a 1 mmol / L decrease in fasting blood glucose, and 15 points deducted for a 1 mmol / L increase); ΔCon is the physical fitness improvement score (5 points for a 0.1 decrease in physical fitness deviation, and 8 points deducted for a 0.1 increase); ΔBMI is the weight improvement score (8 points for a 1 kg / m² decrease in BMI, and 10 points deducted for a 1 kg / m² increase); and Exe is the execution score (average of the three strategy execution scores × 10). The reward range is -50 to 50 points, reflecting the strategy's effectiveness and user compliance in real time.

[0055] 2. Model Principles and Formulas A deep Q-network (DQN) is used to approximate the Q-function (state-action value function) through a neural network to achieve dynamic policy recommendation.

[0056] Q-function (state-action value): , in: Q: The State-Action Value Function represents the expected cumulative reward for performing action a in state s; s: Current state (16-dimensional feature vector), including diabetes risk probability, TCM constitution (one-hot encoding), lifestyle score, and strategy execution degree; a: The current action (one of 27 strategy combinations) is composed of a combination of dietary plans (such as dampness-removing diets), exercise plans (such as jogging), and traditional Chinese medicine conditioning plans (such as moxibustion at Fenglong acupoint); θ: Current Q-network parameters (neural network weights and biases), dynamically updated through training; r: Instant reward (overall effect score), ranging from -50 to 50 points, calculated based on blood sugar improvement, physical condition improvement, weight change and execution rate; γ: Discount factor (set to 0.9 in this scheme), used to weigh the weight of immediate rewards against future rewards; s': The new state after performing action a, including updated risk probability, physical fitness score and lifestyle data; a': Candidate action under the new state s' (one of 27 policy combinations, selected by the target network); Target network parameters (copied from the current network every 100 steps) are used to calculate the target Q-value for stable training; This formula defines the long-term value of a state-action pair, which is the weighted sum of the immediate reward *r* obtained after performing action *a* and the value of the future optimal action. The current network estimates the Q-value in real time, while the target network periodically provides a stable target value. Network parameter optimization is achieved by minimizing the difference between the two.

[0057] Loss function (optimization objective): , Where E is used to calculate the mean of the random variable, which here represents the expected mean square error of the target Q value and the current Q value Q(s,a;θ) of all samples in the experience replay pool; by randomly sampling 32 samples (mini-batch data) from the experience replay pool, the average mean square error of the batch of samples is calculated to approximately estimate the expectation of the overall data distribution, so as to stabilize the training process and reduce the risk of overfitting. Let be the target Q-value (calculated by the target network), and be the Q-value predicted by the current network. Update the current network parameters by minimizing the mean squared error between the target Q-value and the current Q-value. Randomly sample 32 samples from the empirical replay pool, calculate the difference between the target Q-value and the current Q-value, and update the network weights through backpropagation using the Adam optimizer, so that the current Q-value gradually approaches the target Q-value.

[0058] 3. Model Training and Inference Experience Replay Pool: Employs a prioritized experience replay mechanism, storing data in quintuples (s, a, r, s', done) with a capacity of 10,000 records. Sample priority is dynamically adjusted based on the TD error ( ), with samples having a higher probability of being sampled (the sampling probability formula is , where α=0.6 is the priority factor and β=0.4 is the importance sampling weight). Before each training round, the experience replay pool undergoes a sample shuffle to break the temporal correlation between samples and alleviate training oscillations.

[0059] Training parameters: The network structure is a 3-layer fully connected neural network (input layer 16 → hidden layer 64 → hidden layer 32 → output layer 27), with ReLU activation function; the optimizer is Adam (learning rate 0.001, β1=0.9, β2=0.999, ε=1e-8); batch size is 32 (32 samples are sampled from the replay pool each time); the target network update frequency is 100 steps (the current network parameters are copied to the target network every 100 training epochs); the exploration strategy is ε-greedy (ε decays linearly from 1.0 to 0.1, with a decay rate of 0.001 / step); the total number of training steps is 50,000, and the model performance is validated every 1,000 steps (the average Q-value is calculated on the validation set).

[0060] Reasoning process: After inputting state s, the following steps are performed: (1) State preprocessing: normalize the risk probability and lifestyle score (range 0-1), and keep the physical characteristics one-hot encoded; (2) Q-value calculation: input the preprocessed state vector into the current network to obtain the Q-value vector Q(s,a;θ) of 27 actions; (3) Action selection: adopt the ε-greedy strategy (ε is fixed at 0.1 during the reasoning stage), select the action with the largest Q value with a 90% probability, and randomly select the action with a 10% probability (to increase the diversity of strategies); (4) Strategy generation: parse the selected action (a) into a specific strategy combination (e.g., a=15 corresponds to "low GI diet + swimming + acupoint massage"), and add execution suggestions (e.g., diet recipes, exercise duration, moxibustion frequency). The average reasoning time is 0.01 seconds / time, and the strategy recommendation accuracy (the proportion of users with rewards >0 after execution) reaches 78%.

[0061] III. System Interaction Flow like Figure 2 As shown, the interaction process of this system is divided into six major steps: data acquisition, preprocessing, physical condition identification, risk prediction, strategy recommendation, and feedback optimization, as detailed below: (I) Step 1: Data Acquisition and Preprocessing Data collection: Users enter basic information (age, gender, family history) through the app.

[0062] The smartwatch automatically collects activity (steps, distance) and sleep (deep sleep duration) data, transmitting it to the phone via Bluetooth every hour.

[0063] The smart blood glucose meter collects blood glucose data and transmits it to a mobile phone via Bluetooth after measurement.

[0064] Users use a tongue / pulse diagnostic instrument to collect images / pulse data and transmit them to their mobile phones via USB / Bluetooth.

[0065] Data transmission: The mobile app encrypts the data using AES-256 and transmits it to the edge server (community health service center) via SSL protocol. The edge server then forwards real-time data (such as blood glucose) to the cloud server.

[0066] Data preprocessing: Missing value handling: Missing exercise data were filled with the mean of the previous 7 days, and missing blood glucose data were filled with linear interpolation.

[0067] Normalization: blood glucose ( (Normalized to 0-1), number of steps ( ).

[0068] Feature engineering: Extracting tongue coating H value (hue) and tongue entropy (texture complexity) from tongue diagnosis images; extracting pulse frequency (beats / minute) and amplitude (mmHg) from pulse diagnosis data.

[0069] (II) Step 2: Traditional Chinese Medicine Constitution Identification Input data: preprocessed tongue diagnosis features (7 dimensions), pulse diagnosis features (3 dimensions), and body mass scale scores (9 dimensions).

[0070] Model inference: The edge server calls the SVM model and outputs the constitution type (e.g., phlegm-dampness constitution, probability 0.85).

[0071] Results storage: Body constitution type is stored in a MySQL database on a cloud server.

[0072] (III) Step 3: Diabetes Risk Prediction Input data: Preprocessed Western medicine indicators (3-dimensional), traditional Chinese medicine constitution (one-hot, 9-dimensional), family history (1-dimensional), and lifestyle (3-dimensional).

[0073] Model inference: The cloud server calls the XGBoost model and outputs the risk probability (e.g., 0.85).

[0074] Results storage: Risk probabilities are stored in a MySQL database.

[0075] (iv) Step 4: Intelligent Strategy Recommendation Input data: risk probability (1 dimension), TCM constitution (one-hot, 9 dimensions), lifestyle (3 dimensions), current strategy implementation status (3 dimensions).

[0076] Model inference: The cloud server calls the DQN model and outputs the action with the largest Q value (such as "disease-removing diet + jogging + moxibustion at Fenglong acupoint").

[0077] Strategy Display: The cloud server transmits the recommended strategies to the mobile APP, which displays them in the form of pictures and text (e.g., "Diet: Reduce oily food and increase poria and barley porridge; Exercise: Jog for 40 minutes every day; Traditional Chinese Medicine: Moxibustion at Fenglong acupoint for 15 minutes / time").

[0078] (v) Step 5: Feedback and Model Optimization User feedback: Users provide feedback on the implementation status (e.g., "The diet has been followed for 6 days") and the results (e.g., "Blood sugar dropped from 6.5 mmol / L to 5.8 mmol / L") through the APP.

[0079] Reward Calculation: The feedback optimization module calculates rewards based on expert weights (e.g., a 0.7 mmol / L decrease in blood sugar earns 2.8 points, a 2 kg decrease in weight earns 2 points, and the total reward is 9.355 points).

[0080] Sample storage: Store the state (s), action (a), reward (r), and next state (s') into the experience replay pool.

[0081] Model training: When the number of samples in the experience replay pool is ≥32, the cloud server samples 32 samples from the pool and uses the Adam optimizer to adjust the DQN model parameters. Minimize the loss function Update the target network parameters every 100 training steps. ( ).

[0082] System Application Case 1: Taking a user with phlegm-dampness constitution (male, 45 years old, fasting blood glucose 6.5 mmol / L, glycated hemoglobin 6.2%, family history of diabetes) as an example, the application process of the system is demonstrated: (I) Data Acquisition and Preprocessing Data collected by the smartwatch: 8,000 steps, 4 hours of deep sleep.

[0083] Smart blood glucose meter reading: Fasting blood glucose 6.5 mmol / L.

[0084] Tongue diagnosis instrument data: Tongue coating H value (hue) 0.2 (yellowish), tongue texture entropy (texture) 0.8 (rough).

[0085] Pulse diagnosis instrument data: pulse frequency 75 beats / minute, amplitude 15 mmHg (soft and slow pulse).

[0086] After pretreatment: normalized blood glucose value 0.35 ( The normalized value of the number of steps is 0.8 (8000 / 10000=0.8).

[0087] (II) Identification of Traditional Chinese Medicine Constitution Input: Tongue diagnosis features (H=0.2, entropy=0.8), pulse diagnosis features (frequency=75, amplitude=15), body mass scale score (phlegm-dampness constitution score 40 / 80).

[0088] Model inference: The SVM outputs a phlegm-dampness constitution probability of 0.85, confirming the constitution type as phlegm-dampness constitution.

[0089] (III) Diabetes Risk Prediction Input: Fasting blood glucose (0.35), glycated hemoglobin (6.2%), phlegm-dampness constitution (one-hot coding), family history (1), exercise score (0.8).

[0090] Model inference: XGBoost outputs a risk probability of 0.85 (the probability of developing diabetes in the next year is 85%).

[0091] (iv) Intelligent Strategy Recommendation State(s): Risk probability 0.85, Phlegm-dampness constitution (one-hot), Motor score 0.8, Current strategy execution status (0,0,0).

[0092] Model inference: DQN calculates the Q value of all actions. The "dampness-removing diet + jogging + moxibustion at Fenglong acupoint" has the highest Q value (10.5), and this strategy is recommended.

[0093] like Figure 3 The image shown is a schematic diagram of the user recommendation interface.

[0094] (v) Feedback and Model Optimization Feedback from the user one month after implementing the strategy: Implementation status: The diet was followed for 25 out of 28 days (implementation rate 0.89), the exercise was followed for 20 out of 28 days (implementation rate 0.71), and the moxibustion was followed for 10 out of 28 days (implementation rate 0.36).

[0095] Results: Fasting blood glucose decreased from 6.5 mmol / L to 5.8 mmol / L (a decrease of 0.7 mmol / L), weight decreased from 75 kg to 73 kg (a decrease of 2 kg), and the phlegm-dampness constitution score decreased from 40 to 35 (a decrease of 5 points).

[0096] Reward Calculation: Blood sugar decrease: 0.7 × 10 × 0.4 = 2.8 points (weight 0.4).

[0097] Weight loss: 2 × 5 × 0.2 = 2 points (weight 0.2).

[0098] Improvement in physical condition: 5 × 2 × 0.4 = 4 points (weight 0.4).

[0099] Execution status: 0.89×3×0.3 + 0.71×3×0.3 + 0.36×3×0.4 = 0.801 + 0.639 + 0.432 = 1.872 points (weight 0.1).

[0100] Total Rewards point.

[0101] Model optimization: The state (s), action (a), reward (r), and next state (s') (risk probability 0.70, motion score 0.9) are stored in the experience replay pool. When the number of samples reaches 32, the DQN model is trained and the parameters are adjusted. This will improve the accuracy of future recommendations.

[0102] To verify the clinical advantages of this regimen, a controlled trial was designed as follows: I. Experimental Design 1. Study subjects: 200 patients with impaired glucose tolerance were randomly divided into an experimental group (this protocol) and a control group (existing clinical protocol), with 100 patients in each group. There were no statistically significant differences in baseline data between the two groups (P>0.05).

[0103] 2. Intervention methods: Experimental group: The integrated traditional Chinese and Western medicine intelligent recommendation system of this invention was used to dynamically adjust diet, exercise, and traditional Chinese medicine constitution conditioning programs. Control group: Followed existing standard clinical protocols (routine dietary guidance + exercise recommendations). 3. Observation period: 6 months 4. Evaluation indicators: strategy implementation rate, accuracy of diabetes risk prediction, and decrease in fasting blood glucose level. II. Experimental Comparison Process 1. Baseline data collection: Fasting blood glucose, TCM constitution type, lifestyle, and other indicators were measured in both groups of patients at enrollment. 2. Intervention Implementation: The experimental group received personalized strategies and recorded their implementation status via a smart terminal, while the control group followed the traditional paper-based guidance plan. 3. Data Monitoring: Collect blood glucose data monthly, conduct diabetes risk assessments every 3 months, and record the implementation status of the strategy. 4. Statistical Analysis: Data were processed using SPSS 26.0. The t-test was used for comparisons between groups, and P < 0.05 was considered statistically significant. III. Comparison Table of Experimental Results

[0104] Note: Compared with existing clinical protocols, this protocol showed significant advantages in all indicators (P<0.01). Therefore, the intelligent recommendation system and method for diabetes prevention based on Traditional Chinese Medicine (TCM) proposed in this study, through the integration of TCM and Western medicine, dynamic adjustment, and personalized recommendations, addresses the shortcomings of existing diabetes prevention systems. Experiments show that the system achieves a strategy execution rate of 70%, a diabetes risk prediction accuracy of 95%, and a blood glucose reduction of 0.65 mmol / L, significantly outperforming existing systems. This approach provides a new perspective for diabetes prevention and has significant clinical application value.

[0105] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 4 As shown, electronic device 410 may include a first processor 2001.

[0106] Optionally, the electronic device 410 may also include a memory 2002 and a transceiver 2003.

[0107] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0108] The following is combined with Figure 4 A detailed description of each component of electronic device 410 is provided below: The first processor 2001 is the control center of the electronic device 410. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0109] Optionally, the first processor 2001 can perform various functions of the electronic device 410 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0110] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 4 CPU0 and CPU1 are shown in the diagram.

[0111] In a specific implementation, as one example, the electronic device 410 may also include multiple processors, for example... Figure 4 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0112] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0113] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently and be connected via the interface circuit of the electronic device 410. Figure 4 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0114] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0115] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 4 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0116] Optionally, the transceiver 2003 can be integrated with the first processor 2001, or it can exist independently and be connected via the interface circuit of the electronic device 410. Figure 4 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0117] It should be noted that, Figure 4 The structure of the electronic device 410 shown does not constitute a limitation on the router. Actual knowledge structure identification devices may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0118] Furthermore, the technical effects of the electronic device 410 can be referred to in the above method embodiments for the technical effects of the intelligent recommendation method for diabetes prevention based on traditional Chinese medicine, which will not be repeated here.

[0119] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0120] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0121] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0122] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0123] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0124] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0125] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0126] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0127] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0128] 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 network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0129] In addition, the functional units in the various embodiments of the present invention 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.

[0130] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a 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 (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0131] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A traditional Chinese medicine-based diabetes prevention strategy intelligent recommendation system, characterized in that, include: The user terminal is used to collect user data, display recommendation strategies, and receive user feedback, including: The intelligent sensing device is communicatively connected to the user terminal and is used to automatically collect the user's blood glucose data and exercise and sleep data. The traditional Chinese medicine diagnostic and treatment equipment is communicatively connected to the user terminal and is used to collect the user's tongue diagnosis images and pulse diagnosis data. The server is connected to the user terminal and is used to receive and process the user data, blood glucose data, exercise and sleep data, tongue diagnosis images and pulse diagnosis data, and generate and output personalized diabetes prevention strategies based on a preset algorithm model.

2. The system of claim 1, wherein, Also includes: The data acquisition module is used to receive data from the intelligent sensing device and the traditional Chinese medicine diagnosis and treatment device, and to encrypt and preprocess the data. The preprocessing includes at least missing value imputation, data normalization and feature extraction.

3. The system of claim 1 or 2, wherein, The server runs a TCM constitution identification module, which is configured to identify the user's constitution type based on tongue diagnosis features, pulse diagnosis features, and the constitution data filled in by the user, using a support vector machine (SVM) model.

4. The system of claim 3, wherein, The server also runs a diabetes risk prediction module, which is configured to predict the user's risk of developing diabetes using an XGBoost model, based at least on the user's Western medicine indicator data, the constitution type output by the traditional Chinese medicine constitution identification module, family history and lifestyle score.

5. The system as described in claim 4, characterized in that, The server also runs a strategy recommendation module, which is configured to take the state vector consisting of the diabetes risk probability, body type, lifestyle score and current strategy execution status as input, calculate through a deep Q-network (DQN) model, select and output recommended strategy combinations from a preset action space that includes diet, exercise and traditional Chinese medicine conditioning plans.

6. The system of claim 5, wherein, The server also runs a feedback optimization module, which is configured to: receive feedback data on the user's execution of the recommendation strategy and changes in physiological indicators, calculate a reward value based on the feedback data, and use the reward value to train and optimize the deep Q-network (DQN) model.

7. The system as described in claim 1, characterized in that, The intelligent sensing devices include intelligent blood glucose meters and smartwatches; the traditional Chinese medicine diagnostic devices include traditional Chinese medicine tongue diagnosis instruments and traditional Chinese medicine pulse diagnosis instruments.

8. The system of claim 5, wherein, The training process of the Deep Q-Network (DQN) model includes: storing state, action, reward, and next state data in an experience replay pool; sampling data from the experience replay pool and updating the parameters of the current Q-network by minimizing the mean squared error loss function between the target Q-value and the current predicted Q-value; and periodically copying the parameters of the current Q-network to the target network.

9. The system of claim 1, wherein, It also includes a data storage module, which is used to store the user's basic information, Western medicine indicator data, body constitution type, raw data of tongue diagnosis and pulse diagnosis, and intermediate and final results of model output.

10. A method for intelligent recommendation of diabetes prevention strategies based on traditional Chinese medicine, characterized in that, Includes the following steps: S1: Collect and preprocess the user's multi-source data, which includes at least blood glucose and exercise sleep data collected by intelligent sensing devices, tongue diagnosis images and pulse diagnosis data collected by traditional Chinese medicine diagnosis and treatment devices, as well as basic information and lifestyle data input by the user. S2: Based on the preprocessed tongue diagnosis features, pulse diagnosis features, and body mass scale data, the user's body constitution type is determined through a traditional Chinese medicine constitution identification model; S3: At least combine the user's Western medicine indicators, the stated body type, family history and lifestyle score, and calculate the user's probability of developing diabetes through a diabetes risk prediction model; S4: The current state is composed of the probability of developing diabetes, body type, lifestyle score and historical strategy execution, and is input into the strategy recommendation model to output a personalized combination of diabetes prevention strategies from the preset action space. S5: Recommend the combination of prevention strategies to the user, and dynamically optimize the strategy recommendation model based on the user's feedback on the effectiveness of the strategy implementation.

Images