Growth promoting intelligent system based on algorithm iteration task

Abstract

The invention discloses an algorithm iteration task-based growth promotion intelligent system, which relates to the technical field of iteration requirements of intelligent optimization algorithms and comprises a data acquisition and preprocessing module, a feature extraction and selection module, a neural network model design module, a model evaluation and optimization module and an algorithm and result output module. By constructing the efficient neural network model, data can be better fitted, growth prediction can be carried out, exercise prescription parameters are dynamically adjusted according to real-time data and feedback of a user, the mapping relation between environment parameters and exercise efficiency is established through cross-layer correlation analysis, an exercise plan is automatically optimized, and the exercise efficiency is improved. The growth condition of the user is continuously monitored and accurately intervened, and the growth promotion effect and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of iterative requirements of intelligent optimization algorithms, and specifically relates to a growth-promoting intelligent system based on algorithm iterative tasks. Background Art

[0002] Height is an important factor in children's self-confidence and social skills. Monitoring and management can help avoid psychological stress and social barriers caused by height problems. Reasonably arranging study and exercise time according to children's growth potential and development stage helps children achieve a balance between academic performance and physical development. Through regular monitoring, endocrine diseases such as growth retardation, growth hormone deficiency, and hypothyroidism, as well as nutrition-related problems such as malnutrition and obesity, can be detected in a timely manner, enabling early intervention and treatment. Understanding a child's growth status helps parents and doctors develop reasonable diet, exercise, and sleep plans for the child, promoting the child's healthy growth. Monitoring data can provide important information for the child's health record, helping doctors better understand the child's growth trajectory and providing a basis for future health decisions.

[0003] In the prior art, the data source is single, relying only on single biological indicators such as height and bone age, lacking synchronous monitoring of multi-dimensional growth-related factors such as exercise intensity, sleep quality, and hormone levels, and unable to construct a holographic model of the human growth state; using a fixed threshold judgment standard (such as a bone age-height comparison table), no dynamic growth prediction model is established, ignoring the specificity of individual growth curves and unable to adapt to non-linear change stages such as the adolescent growth spurt; the dynamic complexity of the biological growth process leads to model simplification errors, and iterative algorithms are difficult to accurately predict long-term growth trends; there is no self-correction function, the evaluation algorithm does not introduce a machine learning iterative mechanism, and the system misjudgment rate increases with the accumulation of use time; the interaction frequency is unbalanced, requiring multiple manual records per day, but the effective feedback cycle is too long, resulting in the loss of the user's sense of purpose. Summary of the Invention

[0004] To solve the above technical problems, a growth-promoting intelligent system based on algorithm iteration tasks is provided. This technical solution solves the problem of single data source above, relying only on single biological indicators such as height and bone age, lacking synchronous monitoring of multi-dimensional growth-related factors such as exercise intensity, sleep quality, and hormone levels, and being unable to construct a holographic model of the human growth state; using a fixed threshold judgment standard (such as the bone age-height comparison table), not establishing a dynamic growth prediction model, ignoring the specificity of individual growth curves, and being unable to adapt to non-linear change stages such as the adolescent growth spurt; the dynamic complexity of the biological growth process leads to model simplification errors, and the iterative algorithm is difficult to accurately predict long-term growth trends; there is no self-correction function, the evaluation algorithm does not introduce a machine learning iteration mechanism, and the system misjudgment rate increases with the accumulation of usage time; the interaction frequency is unbalanced, requiring multiple manual records per day, but the effective feedback cycle is too long, resulting in the loss of the user's sense of purpose.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A growth-promoting intelligent system based on algorithm iteration tasks, comprising:

[0007] Data acquisition and preprocessing module: Collect multi-source data including biological characteristics, environmental monitoring, and exercise monitoring; preprocess and normalize the collected multi-source data;

[0008] Feature extraction and selection module: Based on the preprocessed data, extract features from the biological characteristics layer, environmental monitoring, and exercise monitoring; use correlation analysis to screen features highly correlated with growth height, and optimize the feature set using the feature importance method;

[0009] Neural network model design module: Determine the dimension of the input layer according to the number of features, and select the number of hidden layers and neurons, and use activation functions; design the output layer according to the task type, add regularization to prevent overfitting, and select the loss function according to the task; select the neural network model architecture;

[0010] Model evaluation and tuning module: Calculate the accuracy, recall rate, and classification tasks of the model, draw learning curves and confusion matrices, and analyze the overfitting and underfitting of the model; use grid search and random search to optimize hyperparameters, and adopt cross-validation to verify the tuning effect;

[0011] Algorithm and result output module: Predict the input data, generate growth promotion suggestions and predicted values; select a suitable iteration strategy according to the system task and data characteristics, draw on the execution-evaluation mechanism in reinforcement learning, evaluate and update the prediction results of the neural network model; display the results through charts and text reports, provide a user interface, and allow users to input new data and obtain real-time predictions.

[0012] Preferably, the data acquisition and preprocessing module specifically includes:

[0013] Data acquisition unit: including a biometric layer, an environmental monitoring layer, and a motion monitoring layer;

[0014] Biometric layer: integrating genetic and physiological data, including congenital growth potential parameters such as gender, parental height, age, and puberty development status;

[0015] Environmental monitoring layer: collecting exogenous growth influencing factors such as environmental light intensity, daily effective sleep duration, and environmental temperature and humidity data through Internet of Things sensors;

[0016] Motion monitoring layer: docking with intelligent sports hardware to record exercise frequency, intensity, and action completion;

[0017] Data preprocessing unit: filling missing values with mean and median. For time series data, interpolation method is used to process missing values; using statistical methods to detect outliers and correct them; converting data from different sources into a unified format and unit; converting non-numerical data into numerical data, using one-hot encoding to process categorical data, and performing word frequency statistics and TF-IDF conversion on text data.

[0018] Preferably, the feature extraction and selection module specifically includes:

[0019] Biometric layer feature extraction unit: statistical features calculate mean, variance, median, and standard deviation to describe the central tendency and distribution of biometric data; calculate skewness and kurtosis to measure the asymmetry and peakedness of data distribution; use correlation analysis to calculate the correlation between biometric parameters and growth height;

[0020] Environmental monitoring layer feature extraction unit: time series feature extraction, calculating mean, variance, median, and standard deviation to reflect the volatility of environmental parameters; calculating the autocorrelation coefficient of the time series of environmental parameters to reflect the periodicity of the data; smoothing the time series data of environmental parameters to extract the long-term trend;

[0021] Motion monitoring layer feature extraction unit: motion feature calculation includes exercise frequency, exercise intensity, and action completion.

[0022] Preferably, the biometric layer feature extraction unit specifically includes:

[0023] Using correlation analysis to calculate the correlation between biometric parameters and growth height, Pearson correlation coefficient:

[0024]

[0025] where r is the Pearson correlation coefficient, Xi is the i-th observed value of the biological characteristic parameter, Y i is the i-th observed value of the growth height, is the mean value of the biological characteristic parameter, is the mean value of the growth height, and n is the number of samples.

[0026] Preferably, the environmental monitoring layer feature extraction unit specifically includes:

[0027] Calculate the autocorrelation coefficient of the time series of environmental parameters:

[0028]

[0029] In the formula, ρ k is the autocorrelation coefficient at lag k, μ is the mean value, X t is the t-th data point of the time series data, k is the lag step, and n is the length of the time series.

[0030] Preferably, the feature extraction and selection module specifically includes:

[0031] Feature selection unit: Filter out significantly relevant features according to the correlation threshold; use random forest and gradient boosting tree models to calculate the importance scores of features, and select the top-ranked features according to the importance scores; use the recursive feature elimination method to gradually screen out the optimal feature subset;

[0032] Feature fusion and optimization unit: Fuse features of different data types to form a comprehensive feature set, use the principal component analysis method to reduce the dimension of the features and reduce the feature dimension; evaluate the performance of the feature set through cross-validation, optimize the feature set according to the prediction performance of the model, and remove redundant features;

[0033] Closed-loop optimization unit: Incremental learning mechanism, online random forest retrains the user's growth response sensitivity regularly, dynamically adjusts the exercise prescription parameters, and correspondingly adjusts the exercise intensity gradient and exercise-rest cycle ratio; Cross-layer association analysis to establish the mapping relationship between environmental parameters and exercise efficacy.

[0034] Preferably, the feature selection unit specifically includes:

[0035] Use random forest and gradient boosting tree models to calculate the importance scores of features, and select the top-ranked features according to the importance scores, including:

[0036]

[0037] In the formula, Importance(j) is the importance score of feature j, T is the number of trees in the random forest, N mwhere \(n\) is the number of samples in node \(m\), \(N\) is the total number of samples, and \(\Delta Gini(m, j)\) is the reduction in Gini coefficient of feature \(j\) at node \(m\);

[0038]

[0039] In the formula, \(Important(L)\) is the importance score of feature \(L\), \(W\) is the number of trees in the gradient boosting tree, and \(SplitCount(m, j)\) is the number of splits of feature \(j\) at node \(m\).

[0040] Preferably, the neural network model design module specifically includes:

[0041] Input layer design: Determine the dimension of the input layer according to the number of features to ensure that the input data matches the dimension of the input layer, and represent it in matrix form, where each row represents a sample and each column represents a feature;

[0042] Hidden layer design: Select the number of hidden layers according to the complexity of the task and the data scale, and determine the number of neurons in each layer based on experience and experiments; Use ReLU and Sigmoid activation functions to introduce non-linearity;

[0043] Output layer design: Determine the dimension of the output layer according to the task type. For regression tasks, the output layer has one neuron for predicting continuous values. For classification tasks, the number of neurons in the output layer is equal to the number of classes, and the Softmax activation function is used for multi-classification and the Sigmoid activation function is used for binary classification; Select the activation function according to the task type, use the linear activation function for regression tasks, and use the Softmax and Sigmoid activation functions for classification tasks; Add L1 and L2 regularization to prevent the weights from being too large; Select the loss function according to the task type, use the mean squared error and mean absolute error for regression tasks, and use the cross-entropy loss function for classification tasks;

[0044] Model architecture selection: Multilayer perceptron is used for structured data tasks; Convolutional neural network is used for image data tasks to extract spatial features through convolutional layers; Recurrent neural network and its variants are used for time series data tasks to capture time dependencies; The Transformer architecture is used for natural language processing tasks to capture long-range dependencies through self-attention mechanisms.

[0045] Preferably, the model evaluation and tuning module specifically includes:

[0046] Model performance evaluation unit: Calculate accuracy and recall, draw learning curves to show the performance of the model on the training set and validation set as the amount of training data and the number of training rounds change, and determine whether the model has overfitting or underfitting; Draw a confusion matrix to show the comparison between the predicted results and the actual results of the model and understand the classification performance of the model;

[0047] Model Tuning Unit: Using grid search to traverse the specified hyperparameter combinations, find the parameter combination that optimizes the model performance, and search for the best model parameters by setting different hyperparameter values; Combining random search to randomly sample in the hyperparameter space, conduct multiple experiments, and find the hyperparameter combination with good performance; Adopting k-fold cross-validation to evaluate the generalization ability and stability of the model.

[0048] Preferably, the algorithm and result output module specifically includes:

[0049] Algorithm Iteration Unit: According to the system tasks and data characteristics, select appropriate iteration strategies, including value iteration, policy iteration, and hybrid iteration strategies; Draw on the execution-evaluation mechanism in reinforcement learning to evaluate and update the prediction results of the neural network model, and use the reward signal in reinforcement learning to guide the model to continuously adjust the prediction strategy to approach the optimal strategy; Combine transfer learning technology to transfer the prior knowledge obtained in previous tasks to new tasks, accelerate the learning process of new tasks, and improve the generalization ability of the algorithm.

[0050] Prediction Unit: Input the preprocessed data into the trained neural network model for growth prediction, and generate prediction values including the predicted growth height and growth trend.

[0051] Result Generation Unit: Based on the prediction results, customize and recommend personalized growth promotion suggestions, including diet adjustment suggestions, exercise plans, and improvement of living habits, based on existing medical research and nutritional knowledge and combined with the user's specific data; Generate a detailed text report, including prediction values, growth trend analysis, and health suggestions.

[0052] User Interaction and Feedback Unit: Provide a user feedback mechanism for the user to evaluate the prediction results and suggestions and ask questions; Collect user feedback for further optimization and improvement of the model.

[0053] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0054] The present invention proposes that through the collection of multi-source data such as biological characteristics, environmental monitoring, and motion monitoring, various factors affecting growth can be comprehensively captured, providing a rich data basis for subsequent feature extraction and model training; through correlation analysis, feature importance assessment, and recursive feature elimination, features significantly related to growth height are screened out, avoiding the interference of redundant features, improving the prediction performance of the model, further optimizing the feature set, and ensuring the efficiency and accuracy of the model; by calculating indicators such as accuracy and recall, drawing learning curves and confusion matrices, the performance of the model is comprehensively evaluated to ensure the stability and reliability of the model; grid search and random search are used to optimize hyperparameters, and the tuning effect is optimized through cross-validation, further improving the prediction performance of the model; through multi-dimensional algorithms such as online random forest, the growth response sensitivity of users is retrained every 24 hours, and the motion prescription parameters are dynamically adjusted, realizing continuous monitoring and precise intervention of the growth status of users; the mapping relationship between environmental parameters and motion efficacy is established, the motion plan is optimized, and the effect and efficiency of growth promotion are improved; through the algorithm and result output module, new data input by users can be predicted in real time, personalized growth promotion suggestions and prediction values are generated, and the results are intuitively displayed in various forms such as charts and text reports, providing a convenient and intuitive growth monitoring and promotion tool for users;

[0055] Through multi-source data fusion, precise feature extraction, efficient model design, and closed-loop optimization algorithms, the scientificity and effectiveness of growth promotion are significantly improved. Brief Description of the Drawings

[0056] Figure 1 It is an internal framework diagram of a growth promotion intelligent system based on algorithm iteration tasks;

[0057] Figure 2 It is an internal composition diagram of the data acquisition and preprocessing module;

[0058] Figure 3 It is an internal composition diagram of the feature extraction and selection module. Detailed Embodiments

[0059] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments in the following description are only examples, and those skilled in the art can think of other obvious variations.

[0060] Referring to Figure 1 As shown, a growth promotion intelligent system based on algorithm iteration tasks includes:

[0061] Data acquisition and preprocessing module: Collect multi-source data including biological characteristics, environmental monitoring, and motion monitoring; preprocess and normalize the collected multi-source data;

[0062] Feature Extraction and Selection Module: Based on the preprocessed data, extract features from the biological characteristic layer, environmental monitoring, and motion monitoring; use correlation analysis to screen features related to growth height, and adopt feature importance methods to optimize the feature set;

[0063] Neural Network Model Design Module: Determine the dimension of the input layer according to the number of features, and select the number of hidden layers and neurons, and use activation functions; design the output layer according to the task type, add regularization to prevent overfitting, and select the loss function according to the task; select the neural network model architecture;

[0064] Model Evaluation and Tuning Module: Calculate the accuracy, recall rate, and classification tasks of the model, draw learning curves and confusion matrices, and analyze the overfitting and underfitting of the model; use grid search and random search to optimize hyperparameters, and adopt cross-validation to verify the tuning effect;

[0065] Algorithm and Result Output Module: According to the system tasks and data characteristics, select appropriate iterative strategies, draw on the execution-evaluation mechanism in reinforcement learning to evaluate and update the prediction results of the neural network model; predict the input data, generate suggestions for growth promotion and predicted values; display the results through charts and text reports, provide a user interface, and allow users to input new data and obtain real-time predictions.

[0066] It should be noted that this invention provides the first medical evidence-based closed-loop control system in the field of intelligent growth promotion, with significant technological advancement and clinical practical value. Tasks are directly issued according to the background data algorithm. Users can view their training reports and effects in the App. For example, users only need to complete the preset high-touch training according to the background system.

[0067] The communication architecture between the cloud server and the hardware end of this intelligent system includes:

[0068] Mobile-Phone-Free Wi-Fi Direct Connection: The intelligent hardware is built-in with a Wi-Fi module (supporting the 802.11ac protocol) to directly establish two-way communication with the cloud server, avoiding the relay dependence on mobile phones in the traditional Bluetooth solution and reducing the frequency of children using mobile terminals; the high-touch training tasks are transmitted to the high-touch device through Wi-Fi, and the training tasks are also transmitted to the server through Wi-Fi, separating the training device and the App, allowing children users to only use the high-touch training device, and parents can view the children's training in the App; the algorithm obtains the training effect, and parents can also re-guide the children's training in the App according to the training evaluation.

[0069] Intelligent Control of the Hardware End, Offline Task Execution, and the Cloud Sends Encrypted Task Packages to the Local Storage of the Hardware, Supporting Operation According to Preset Programs in a Disconnected Environment;

[0070] Generating growth promotion suggestions and linking with the hardware side includes: The hardware side can realize real-time display of the high jump target line through laser projection, and the vibration prompt frequency of the adaptive skipping rope handle can be adjusted (triggered when the deviation exceeds the set value of ±5%);

[0071] Medical compliance design of the hardware side:

[0072] Safety threshold control: Build a knowledge base of sports medicine experts. When the exercise volume exceeds the threshold corresponding to the bone age (for example, the daily jump volume of children before puberty > 300 times), the hardware side will be forced to enter the protection mode;

[0073] Growth warning system: Automatically generate a medical consultation and recommendation letter based on the monthly growth rate (such as < 0.4 cm / month) and push it to the bound guardian terminal.

[0074] The strategies for algorithm iteration in the algorithm and result output module include:

[0075] Select appropriate iteration strategies, including value iteration, policy iteration, or hybrid iteration strategies; Value iteration is suitable for offline learning scenarios, policy iteration is suitable for online learning scenarios, and the hybrid iteration strategy combines the advantages of both and can be flexibly selected according to actual needs;

[0076] Draw on the execution-evaluation mechanism in reinforcement learning to evaluate and update the prediction results of the neural network model. Use the reward signal in reinforcement learning to guide the model to continuously adjust the prediction strategy to approach the optimal strategy; During the iteration process, adopt acceleration strategies such as adjustable value iteration, hybrid iteration, etc. to reduce the number of iterations and computational complexity, and improve the convergence speed and efficiency of the algorithm; At the same time, by introducing techniques such as auxiliary variables and experience replay, improve the data utilization rate and enhance the stability and robustness of the algorithm;

[0077] Combine transfer learning techniques to transfer the prior knowledge obtained in previous tasks to new tasks, accelerate the learning process of new tasks, and improve the generalization ability of the algorithm. For example, use historical data to build a model network, and use the cost function obtained from model-based evaluation learning as prior knowledge to guide the learning of new tasks; After each iteration, evaluate the performance metrics of the model, such as accuracy, recall rate, overfitting and underfitting conditions, etc.; According to the evaluation results, adjust the iteration strategy, hyperparameters, or neural network model architecture to further optimize the model performance;

[0078] The above-mentioned policy iteration includes policy evaluation and policy improvement:

[0079] Policy evaluation calculates the state value function through the Bellman expectation equation:

[0080]

[0081] In the formula, V π(s) is the state value function, representing the expected cumulative reward when in state s under policy π; π(a∣s) is the probability of selecting action a in state s; R(s,a) is the immediate reward, representing the immediate reward obtained by executing action a in state s; γ is the discount factor, with a value range of 0 < γ ≤ 1, used to measure the current value of future rewards; P(s'|s,a) is the state transition probability, representing the probability of transitioning to state s′ after executing action a in state s; V π (s') is the value function of the next state s′, representing the expected cumulative reward when in state s′ under policy π;

[0082] Update the policy to maximize the expected return:

[0083]

[0084] In the formula, π k+1 (a|s) is the updated policy, representing the probability of selecting action a in state s; a is the action, representing the actions that can be executed in state s; s is the state, representing the current environmental state; argmax a Q π (s,a) is to select the action a that maximizes the action value function Q π (s,a); Q π (s,a) is the action value function of state s and action a under policy π.

[0085] By continuously adjusting and optimizing the model through policy iteration, the model can more accurately predict the growth situation, generate more reliable growth promotion suggestions. Based on more accurate prediction results, the system can provide a more reasonable and effective growth promotion plan for users; the application of transfer learning technology enables the system to utilize the prior knowledge obtained in previous tasks, accelerate the learning process of new tasks, and reduce the time and resources required for learning from scratch; evaluating the model performance after each iteration and adjusting the iteration strategy, hyperparameters, or neural network model architecture according to the evaluation results can effectively reduce the risk of model overfitting and improve the generalization ability of the system; the algorithm iteration strategy enables the system to continuously adjust and optimize its own strategy according to new data and feedback, realizing the continuous improvement and upgrade of the system.

[0086] Referring to Figure 2 as shown, the data acquisition and preprocessing module includes:

[0087] Data acquisition unit: including a biological characteristics layer, an environmental monitoring layer, and a motion monitoring layer;

[0088] Biological characteristics layer: integrating genetic and physiological data, including congenital growth potential parameters such as gender, parental height, age, and puberty development status;

[0089] Environmental monitoring layer: Collect exogenous growth influencing factors such as environmental light intensity, daily effective sleep duration, and environmental temperature and humidity data through Internet of Things sensors;

[0090] Sports monitoring layer: Connect to intelligent sports hardware to record exercise frequency, intensity, and action completion;

[0091] Data preprocessing unit: Use mean and median to fill in missing values. For time series data, use interpolation method to handle missing values; Detect outliers using statistical methods and correct them; Convert data from different sources into a unified format and unit; Convert non-numerical data into numerical data, use one-hot encoding to process categorical data, and perform word frequency statistics and TF-IDF conversion on text data.

[0092] It should be noted that the biological characteristics layer is mainly responsible for integrating genetic and physiological data, which are important indicators for evaluating an individual's growth potential, including: Gender has a significant impact on growth and development. For example, there are differences in growth rates and final heights between males and females during puberty; Genetic factors play an important role in determining height, and the heights of parents can be used as important references for predicting the heights of their children; Age is a key factor in growth and development, and the growth rates and needs are different at different ages; Puberty is a critical period for growth and development, and the development status directly affects an individual's growth potential;

[0093] The environmental monitoring layer collects data on exogenous growth influencing factors through Internet of Things sensors, which have important impacts on growth and development, including: Light intensity affects biological rhythms and the synthesis of vitamin D, thus affecting growth and development; Sleep is crucial for the secretion of growth hormone, and sufficient sleep helps growth; Temperature and humidity affect comfort and metabolism, thus affecting the growth environment;

[0094] The sports intervention layer connects to intelligent sports hardware to record the user's sports data, which are used to evaluate the promoting effect of exercise on growth, including: The number of times of exercise per week, reflecting the regularity of exercise; The intensity level of exercise, such as low, medium, and high, reflecting the load of exercise; The completion quality of exercise actions, such as completion rate and standardization, reflecting the effect of exercise.

[0095] Refer to Figure 3 As shown, the feature extraction and selection module includes:

[0096] Biological characteristics layer feature extraction unit: Statistical features describe the central tendency and distribution of biological characteristics data by calculating mean, variance, median, and standard deviation; Calculate skewness and kurtosis to measure the asymmetry and peakedness of data distribution; Use correlation analysis to calculate the correlation between biological characteristics parameters and growth height;

[0097] Environmental Monitoring Layer Feature Extraction Unit: Extract time series features, calculate the mean, variance, median, and standard deviation to reflect the volatility of environmental parameters; calculate the autocorrelation coefficient of the time series of environmental parameters to reflect the periodicity of the data; smooth the time series data of environmental parameters to extract the long-term trend;

[0098] Motion Monitoring Layer Feature Extraction Unit: Motion feature calculation includes motion frequency, motion intensity, and action completion.

[0099] The biological characteristic layer feature extraction unit specifically includes:

[0100] Use correlation analysis to calculate the correlation between biological characteristic parameters and growth height, Pearson correlation coefficient:

[0101]

[0102] In the formula, r is the Pearson correlation coefficient, X i is the i-th observation value of the biological characteristic parameter, Y i is the i-th observation value of the growth height, is the mean of the biological characteristic parameters, is the mean of the growth height, and n is the number of samples.

[0103] The environmental monitoring layer feature extraction unit specifically includes:

[0104] Calculate the autocorrelation coefficient of the time series of environmental parameters:

[0105]

[0106] In the formula, ρ k is the autocorrelation coefficient at lag k, μ is the mean, X t is the t-th data point of the time series data, k is the lag step, and n is the length of the time series.

[0107] The feature extraction and selection module specifically includes:

[0108] Feature Selection Unit: According to the correlation threshold, screen out significantly correlated features; use random forest and gradient boosting tree models to calculate the importance scores of features, and select the top-ranked features according to the importance scores; use the recursive feature elimination method to gradually screen out the optimal feature subset;

[0109] Feature Fusion and Optimization Unit: Fuse features of different data types to form a comprehensive feature set, use the principal component analysis method to reduce the dimension of the features and reduce the feature dimension; evaluate the performance of the feature set through cross-validation, and optimize the feature set according to the prediction performance of the model to remove redundant features;

[0110] Closed-loop optimization unit: Incremental learning mechanism. The online random forest retrains the user's growth response sensitivity regularly, dynamically adjusts the exercise prescription parameters, and correspondingly adjusts the exercise intensity gradient and the exercise-rest cycle ratio; Cross-layer correlation analysis is performed to establish the mapping relationship between environmental parameters and exercise efficacy.

[0111] The feature selection unit specifically includes:

[0112] Using random forest and gradient boosting tree models, calculate the importance scores of features, and select the top-ranked features according to the importance scores, including:

[0113]

[0114] In the formula, Importance(j) is the importance score of feature j, T is the number of trees in the random forest, N m is the number of samples in node m, N is the total number of samples, and ΔGini(m,j) is the reduction in Gini coefficient of feature j at node m;

[0115]

[0116] In the formula, Important(L) is the importance score of feature L, W is the number of trees in the gradient boosting tree, and SplitCount(m,j) is the number of splits of feature j at node m.

[0117] It should be noted that the statistical feature calculations in the biological characteristic layer feature extraction unit: mean, variance, median, standard deviation. These statistics are used to describe the central tendency and distribution of biological characteristic data. For example, the mean reflects the average level of the data, the variance and standard deviation reflect the degree of dispersion of the data, and the median reflects the middle position of the data;

[0118] Skewness, kurtosis: Skewness measures the asymmetry of the data distribution. Positive skewness indicates that the data is skewed to the right, and negative skewness indicates that the data is skewed to the left; Kurtosis measures the peakedness of the data distribution. High kurtosis indicates that the data distribution is more concentrated, and low kurtosis indicates that the data distribution is more dispersed;

[0119] Pearson correlation coefficient: Calculate the correlation between biological characteristic parameters (such as gender, parental height, age, etc.) and growth height, and screen out the features that are significantly correlated with growth height.

[0120] The environmental monitoring layer feature extraction unit mainly extracts time series features from environmental monitoring data to reflect the impact of environmental parameters on growth, including:

[0121] Time series feature extraction: mean, variance, median, standard deviation. These statistics are used to describe the volatility of environmental parameters (such as light intensity, sleep duration, temperature and humidity);

[0122] Autocorrelation coefficient: Calculate the autocorrelation coefficient of the time series of environmental parameters to reflect the periodicity of the data. For example, light intensity may have a daily periodicity, and sleep duration may have a weekly periodicity;

[0123] Smoothing processing: Smooth the time series data of environmental parameters by methods such as moving average to extract the long-term trend and remove the influence of short-term fluctuations.

[0124] The dynamic adjustment in the closed-loop optimization unit includes:

[0125] Exercise prescription parameters: Exercise intensity gradient (such as the weekly increase in the target height of touching the ceiling is 0.3 - 1.2 cm), exercise-rest cycle ratio (optimized to 1:0.35 ± 0.05 based on sleep quality data);

[0126] Cross-layer correlation analysis: Establish a mapping relationship between environmental parameters and exercise efficacy. For example, when the environmental temperature > 28°C, automatically reduce the exercise intensity by 15% and extend the intermittent duration; when the sleep duration < 7 hours, trigger the exercise plan pause mechanism.

[0127] In summary, the advantages of the present invention are as follows: Through the feature extraction and selection module, key features in the data are deeply mined. Using methods such as correlation analysis, feature importance evaluation, and recursive feature elimination, features significantly related to growth height are screened out, improving the accuracy and efficiency of the model; The neural network model design module reasonably determines the input layer dimension, the number of hidden layers and neurons according to the number of features and the type of task, and selects appropriate activation functions and loss functions. At the same time, regularization is added to prevent overfitting, constructing an efficient neural network model that can better fit the data and perform growth prediction; The algorithm and result output module can perform real-time prediction based on the user's input data, generate personalized growth promotion suggestions and prediction values, and intuitively display the results in various forms such as charts and text reports, providing a convenient and intuitive growth monitoring and promotion tool for users; Combining the closed-loop optimization algorithm, the system can dynamically adjust exercise prescription parameters according to the user's real-time data and feedback, and establish a mapping relationship between environmental parameters and exercise efficacy through cross-layer correlation analysis, automatically optimizing the exercise plan, realizing continuous monitoring and precise intervention of the user's growth status, and improving the effect and efficiency of growth promotion.

[0128] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A growth-promoting intelligent system based on algorithm-iterative tasks, characterized in that, Including: Data acquisition and preprocessing module: Collect multi-source data including biometric characteristics, environmental monitoring, and motion monitoring; Preprocess and normalize the collected multi-source data; Feature extraction and selection module: Based on the preprocessed data, extract features from the biometric characteristics layer, environmental monitoring, and motion monitoring; Use correlation analysis to screen features related to growth height and optimize the feature set using feature importance methods; Neural network model design module: Determine the dimension of the input layer according to the number of features, select the number of hidden layers and neurons, and use activation functions; Design the output layer according to the task type, add regularization to prevent overfitting, and select the loss function according to the task; Select the neural network model architecture; Model evaluation and tuning module: Calculate the accuracy, recall rate, and classification tasks of the model, draw learning curves and confusion matrices, and analyze the overfitting and underfitting situations of the model; Use grid search and random search to optimize hyperparameters and adopt cross-validation to verify the tuning effect; Algorithm and result output module: According to the system task and data characteristics, select an appropriate iterative strategy, draw on the execution-evaluation mechanism in reinforcement learning to evaluate and update the prediction results of the neural network model; Predict the input data, generate growth promotion suggestions and prediction values; Display the results through charts and text reports, provide a user interface, and allow users to input new data and obtain real-time predictions.

2. The intelligent growth promotion system based on algorithm iterative tasks according to claim 1, characterized in that The data acquisition and preprocessing module specifically includes: Data acquisition unit: including biometric characteristics layer, environmental monitoring layer, and motion monitoring layer; Biometric characteristics layer: Integrate genetic and physiological data, including congenital growth potential parameters such as gender, parental height, age, and puberty development status; Environmental monitoring layer: Collect external growth influencing factors such as environmental light intensity, daily effective sleep duration, and environmental temperature and humidity data through Internet of Things sensors; Motion monitoring layer: Connect to intelligent sports hardware to record exercise frequency, intensity, and action completion; Data preprocessing unit: Use mean and median to fill in missing values. For time series data, use interpolation method to process missing values; Use statistical methods to detect outliers and correct them; Convert data from different sources into a unified format and unit; Convert non-numerical data into numerical data, use one-hot encoding to process categorical data, and perform word frequency statistics and TF-IDF conversion on text data.

3. The promoting growth intelligent system based on algorithm iterative tasks according to claim 2, wherein, The feature extraction and selection module specifically includes: Biometric characteristics layer feature extraction unit: Statistical features describe the central tendency and distribution of biometric data by calculating mean, variance, median, and standard deviation; Calculate skewness and kurtosis to measure the asymmetry and peakedness of data distribution; Use correlation analysis to calculate the correlation between biometric parameters and growth height; Environmental monitoring layer feature extraction unit: Time series feature extraction, calculate mean, variance, median, and standard deviation to reflect the volatility of environmental parameters; Calculate the autocorrelation coefficient of the time series of environmental parameters to reflect the periodicity of the data; Smooth the time series data of environmental parameters to extract the long-term trend; Motion Monitoring Layer Feature Extraction Unit: Motion feature calculation includes motion frequency, motion intensity, and action completion degree.

4. The promoting growth intelligent system based on algorithm iterative tasks according to claim 3, characterized in that, The biological characteristic layer feature extraction unit specifically includes: Using correlation analysis to calculate the correlation between biological characteristic parameters and growth height, Pearson correlation coefficient: where r is the Pearson correlation coefficient, X i is the i-th observed value of the biological characteristic parameter, Y i is the i-th observed value of the growth height, is the mean of the biological characteristic parameter, is the mean of the growth height, and n is the sample size.

5. The promoting growth intelligent system based on algorithm iterative tasks according to claim 3, wherein The environmental monitoring layer feature extraction unit specifically includes: Calculating the autocorrelation coefficient of the time series of environmental parameters: where ρ k is the autocorrelation coefficient at lag k, μ is the mean, X t is the t-th data point of the time series data, k is the lag step, and n is the length of the time series.

6. The growth-promoting intelligent system based on algorithm iterative tasks according to claim 5, wherein The feature extraction and selection module specifically includes: Feature Selection Unit: According to the correlation threshold, screen out significantly relevant features; use the random forest and gradient boosting tree models to calculate the importance scores of features, and select the top-ranked features according to the importance scores; use the recursive feature elimination method to gradually screen out the optimal feature subset; Feature Fusion and Optimization Unit: Fuse features of different data types to form a comprehensive feature set, use the principal component analysis method to reduce the dimension of features and reduce the feature dimension; evaluate the performance of the feature set through cross-validation, optimize the feature set according to the prediction performance of the model, and remove redundant features; Closed-loop Optimization Unit: Incremental learning mechanism, the online random forest retrains the user's growth response sensitivity regularly, dynamically adjusts the exercise prescription parameters, and correspondingly adjusts the exercise intensity gradient and exercise-rest cycle ratio; cross-layer correlation analysis to establish the mapping relationship between environmental parameters and exercise efficacy.

7. An intelligent growth-promoting system based on algorithm-iterative tasks according to claim 6, characterized in that, The feature selection unit specifically includes: Using the random forest and gradient boosting tree models to calculate the importance scores of features, and selecting the top-ranked features according to the importance scores, including: Where Importance(j) is the importance score of feature j, T is the number of trees in the random forest, N m is the number of samples in node m, N is the total number of samples, and ΔGini(m,j) is the decrease in the Gini coefficient of feature j at node m; In the formula, Important(L) is the importance score of feature L, W is the number of trees in the gradient boosting tree, and SplitCount(m,j) is the number of splits of feature j at node m.

8. An intelligent growth promotion system based on algorithm iterative tasks according to claim 7, characterized in that, The neural network model design module specifically includes: Input layer design: Determine the dimension of the input layer according to the number of features, ensure that the input data matches the dimension of the input layer, and represent it in matrix form, where each row represents a sample and each column represents a feature; Hidden layer design: Select the number of hidden layers according to the complexity of the task and the data scale, and determine the number of neurons in each layer based on experience and experiments; use the ReLU and Sigmoid activation functions to introduce non-linearity; Output layer design: Determine the dimension of the output layer according to the task type. Based on the regression task, there is one neuron in the output layer for predicting continuous values. Based on the classification task, the number of neurons in the output layer is equal to the number of categories. Use the Softmax activation function for multi-classification and the Sigmoid activation function for binary classification; select the activation function according to the task type, use the linear activation function for the regression task, and use the Softmax and Sigmoid activation functions for the classification task; add L1 and L2 regularization to prevent the weights from being too large; select the loss function according to the task type, use the mean squared error and mean absolute error for the regression task, and use the cross-entropy loss function for the classification task; Model architecture selection: Multilayer perceptrons are used for structured data tasks; convolutional neural networks are used for image data tasks to extract spatial features through convolutional layers; recurrent neural networks and their variants are used for time series data tasks to capture temporal dependencies; the Transformer architecture is used for natural language processing tasks to capture long-range dependencies through self-attention mechanisms.

9. The promoting growth intelligent system based on an algorithm iterative task according to claim 8, wherein, The model evaluation and tuning module specifically includes: Model performance evaluation unit: Calculate accuracy and recall, plot learning curves to show the performance of the model on the training set and validation set as it changes with the amount of training data and the number of training epochs, and determine whether the model is overfitting or underfitting; plot confusion matrices to show the comparison between the predicted results and the actual results of the model and understand the classification performance of the model. Model tuning unit: Use grid search to traverse the specified hyperparameter combinations to find the parameter combination that optimizes the model performance. By setting different hyperparameter values, search for the best model parameters; combine random search to randomly sample in the hyperparameter space and conduct multiple trials to find good hyperparameter combinations; adopt k-fold cross-validation to evaluate the generalization ability and stability of the model.

10. A growth-promoting intelligent system based on algorithm iteration tasks according to claim 9, characterized in that, The algorithm and result output module specifically includes: Algorithm iteration unit: According to the system tasks and data characteristics, select appropriate iteration strategies, including value iteration, policy iteration, and hybrid iteration strategies; draw on the execution-evaluation mechanism in reinforcement learning to evaluate and update the prediction results of the neural network model, and use the reward signal in reinforcement learning to guide the model to continuously adjust the prediction strategy to approach the optimal strategy; combine transfer learning techniques to transfer the prior knowledge obtained in previous tasks to new tasks, accelerate the learning process of new tasks, and improve the generalization ability of the algorithm. Prediction unit: Input the preprocessed data into the trained neural network model for growth prediction, generating prediction values including predicted growth height and growth trend. Result generation unit: Based on the prediction results, customize and recommend personalized growth promotion suggestions, including dietary adjustment suggestions, exercise plans, and lifestyle improvement, by combining existing medical research and nutritional knowledge with the user's specific data; generate a detailed text report, including prediction values, growth trend analysis, and health suggestions. User interaction and feedback unit: Provide a user feedback mechanism for users to evaluate the prediction results and suggestions and ask questions; collect user feedback for further optimization and improvement of the model.

Images