A method for monitoring children's health based on vital signs data

Abstract

This invention proposes a method for monitoring children's health based on vital sign data, relating to the field of children's health monitoring. The proposed children's health monitoring process includes constructing a children's vital sign dataset; mapping the data sequence at each time step in the dataset to a high-dimensional space using a lookup table; extracting feature sequences using linear transformation and ReLU activation function; dynamically weighting and smoothing the feature sequences using a weighted and smoothing coefficient matrix generated by nonlinear adjustment to obtain a weighted feature sequence; introducing a growth cycle factor to obtain a physiological feature sequence; then using a long short-term memory network to capture a temporal feature sequence containing historical health information; calculating the differences between physiological feature sequences at adjacent time steps; calculating feature reconstruction parameters based on the difference values ​​to generate a health feature sequence; finally, flattening the health feature sequence and inputting it into a fully connected neural network; and outputting children's health monitoring values ​​after activation function and linear transformation, thereby improving the accuracy of children's health monitoring.

Description

Technical Field

[0001] This invention belongs to the field of children's health monitoring, and specifically relates to a method for children's health monitoring based on vital sign data. Background Technology

[0002] Child health monitoring refers to the process of collecting multi-dimensional data on children's physiological and psychological characteristics in real time or periodically, and then analyzing and evaluating these data using modern technology to provide personalized health management recommendations. With the development of the social economy and the increasing number of children's health problems, traditional health management methods are no longer sufficient to meet the needs for accurate monitoring of children's health. Therefore, establishing an accurate and real-time child health monitoring system can help parents and medical staff understand children's health status in real time, promptly identify potential health risks, and provide important evidence for early intervention and health management of children.

[0003] Existing methods for monitoring children's health mainly rely on traditional physical examinations and manual analysis, which have problems such as delayed data collection, long monitoring cycles, and insufficient consideration of individual differences. Traditional physical examinations often only provide children's health information at a certain moment, making it difficult to achieve dynamic tracking and timely response to health status. In addition, the data processing methods in existing technologies mostly rely on simple physiological indicator monitoring, failing to make full use of multi-dimensional vital sign data and lacking accurate modeling of children's personalized health needs, resulting in insufficient accuracy and timeliness of monitoring results.

[0004] Changes in children's health status are driven by multidimensional vital signs data. These data not only exhibit complex nonlinear fluctuations over time, but also present dynamic evolution patterns across multiple scales and stages. By employing children's health monitoring methods that integrate multidimensional vital signs data and historical characteristics, we can systematically depict the long-term trends of children's overall health, thereby achieving precise monitoring of children's health. Summary of the Invention

[0005] This invention provides a method for monitoring children's health based on vital sign data. It assembles all vital sign data of a child at each moment into a data sequence, obtains an implicit feature sequence through a lookup table, and then obtains a feature sequence through linear transformation and ReLU activation function. Nonlinear adjustment is used to generate the weight of the current moment, combined with a smoothing coefficient matrix to generate a weighted feature sequence. A growth cycle factor is further introduced to obtain a physiological feature sequence. A long short-term memory network is used to obtain a temporal feature sequence. The difference between the current and previous physiological feature sequences is used to generate feature reconstruction parameters, achieving adaptive fusion of physiological and temporal feature sequences to form a health feature sequence. The health feature sequence is flattened and input into a fully connected neural network and activation function, and after linear transformation, the final child health monitoring value is output, achieving comprehensive and accurate monitoring of children's health.

[0006] The technical method adopted by the present invention to achieve the above objectives specifically includes the following steps:

[0007] S1. Collect vital signs data that affect children's health, preprocess them using the Z-score method, and construct a children's vital signs dataset;

[0008] S2. All data at each moment in the children's vital signs dataset are combined into a data sequence. A lookup table is used to map the data sequence to a high-dimensional space to obtain an implicit feature sequence reflecting the children's health status. The feature sequence is obtained through linear transformation and ReLU activation function.

[0009] S3. Construct an adaptive temporal weighting module. Based on the feature sequence and the weight of the previous time step, generate the weight of the current time step using nonlinear adjustment. At the same time, calculate the smoothing coefficient matrix, dynamically weight and smooth the feature sequence to obtain the weighted feature sequence.

[0010] S4. Construct a growth cycle feature module, introduce a growth cycle factor, and multiply it element-wise with the weighted feature sequence to obtain a physiological feature sequence. The physiological feature sequence forms time series data, and a long short-term memory network is used to obtain a time series feature sequence.

[0011] S5. Construct a health status fluctuation perception module. By calculating the difference between the physiological feature sequence at the current moment and the previous moment, obtain the health status fluctuation value. Introduce a feature reconstruction parameter. Based on this parameter, weightedly fuse the time series feature sequence and the physiological feature sequence to form a health feature sequence.

[0012] S6. Flatten the health feature sequence into a feature vector, input it sequentially into a fully connected neural network and an activation function, and then obtain the final child health monitoring value through linear transformation.

[0013] Preferably, in step S1, vital sign data affecting children's health are collected, including weight, height, body temperature, heart rate, blood oxygen saturation, ambient temperature, steps, and sleep duration. For each type of data, the Z-score method is used to detect outliers. Any data value with a Z-score greater than 3 or less than -3 is judged as an outlier. For the detected outliers, the mean replacement method is used to process them, that is, the outlier is replaced with the mean of the data three days before and after it, so as to ensure the accuracy and consistency of the data and construct a children's vital sign dataset.

[0014] Preferably, children's vital signs data are often simple numerical stacks of various types, which make it difficult to reveal the complex relationships between different data and cannot fully reflect the potential changes in health status. This invention automatically maps all children's vital signs data to a high-dimensional space through a lookup table, thereby enriching the semantic expression and intrinsic connection of each data item, enabling the model to better understand the deep-seated correlations between various data items. On this basis, linear transformation and nonlinear activation mechanisms are further introduced to reorganize and enhance these high-dimensional implicit features, solving the problem of weak expressive ability of original children's vital signs data and difficulty in capturing health information.

[0015] Preferably, in step S2, in the children's vital signs dataset, all data at each moment constitute the data sequence input to the model, for the first... A sequence of vital signs that constantly affect children's health ,Will The mapping to a richer feature sequence includes the following steps:

[0016] S21. Sequence of vital signs data By mapping a lookup table to a high-dimensional space, a high-dimensional implicit feature sequence is obtained, enabling the model to automatically extract potential health information. The specific mathematical model is as follows:

[0017] ;

[0018] In the formula, For the first Implicit feature sequences that constantly influence children's health data For lookup table functions;

[0019] S22. Using linear transformation to transform implicit feature sequences Mapping to the physiological feature space allows the output features to better fit the actual physiological patterns of children's health, and the ReLU activation function enhances the model's ability to express nonlinear features. The specific mathematical model is as follows:

[0020] ;

[0021] In the formula, For the transformed first Feature sequences that constantly influence children's health data A learnable weight matrix that maps implicit feature sequences to a physiological feature space. This is a learnable bias term.

[0022] Preferably, the present invention maps children's vital sign data sequences into high-dimensional implicit features, which can effectively mine potential health information in the data. The lookup table mapping method introduces more discriminative feature expression capabilities to the model, which is conducive to revealing the complex internal relationships between the original children's vital sign data. By using linear transformation and ReLU activation function, high-dimensional implicit features are automatically extracted and recombined, thereby improving the model's ability to represent nonlinear health information.

[0023] Preferably, children's vital sign data sequences are highly dynamic and noisy. Traditional methods typically use fixed weights or simple moving averages to process the feature sequences, making it difficult to simultaneously consider both historical trends and sudden anomalies in health status. This invention designs an adaptive temporal weighting module that integrates the weights from the previous moment and the information from the feature sequence at the current moment to generate the weights for the current moment. This allows the weights to retain historical information while adjusting promptly based on current data mutations. Secondly, a smoothing coefficient matrix is ​​calculated to suppress high-frequency fluctuations in real time, reducing the interference of short-term noise on the overall health trend assessment. Finally, the feature sequences are weighted and corrected under the combined effect of dynamic weights and the smoothing coefficient matrix, which fully preserves important temporal dynamic information while preventing false alarms and misjudgments, making the model's depiction of children's true health status more accurate.

[0024] Preferably, in step S3, constructing the adaptive temporal weight module specifically includes the following steps:

[0025] S31, Comprehensive Section Weight of time and the first The information of the time-series features is combined with nonlinear adjustment to control the rate and magnitude of weight changes. The specific mathematical model is as follows:

[0026] ;

[0027] In the formula, For the first Weights of the time-series feature sequences For the first Weights of the time-series feature sequences The step size parameter is used to adjust the magnitude of the learnable weight updates. These are learnable weight suppression parameters that balance the influence of the current feature and the previous weight. For learnable nonlinear modulation parameters, adjust the weights to adjust their sensitivity to feature changes.

[0028] S32. Calculate the smoothing coefficient matrix. The size of the smoothing coefficient matrix is ​​automatically adjusted based on the magnitude of change in the feature sequence at adjacent time points to correct the rate of change in children's health data characteristics and suppress noise interference with the fluctuations of the feature sequence. The specific mathematical model is as follows:

[0029] ;

[0030] In the formula, For the first The smoothing coefficient matrix at time step, For the first Feature sequences that constantly influence children's health data For learnable feature changes sensitive parameters, This is for taking the absolute value;

[0031] S33, the first Time-smoothing coefficient matrix Weight With the Feature sequences of children's health data at any time Multiplying them yields a weighted feature sequence. The specific mathematical model is as follows:

[0032] ;

[0033] In the formula, For the first The weighted feature sequence at time step.

[0034] Preferably, the present invention introduces a dynamic weight and smoothing coefficient matrix to sense and adjust changes in vital sign data in real time. The weight update mathematical model integrates historical weights and current feature sequences and introduces nonlinear modulation to enhance the model's adaptability to sudden changes, anomalies, and trend changes. The adaptive adjustment of the smoothing coefficient matrix further suppresses the interference of noise on feature fluctuations, preventing false alarms and misjudgments. After dynamic weighting and smoothing, the feature sequence can more accurately reflect the changing trend of children's true health status.

[0035] Preferably, traditional linear feature processing methods are insufficient to fully reflect the potential physiological patterns and periodic disturbances in children's growth and development, resulting in limited ability of the model to identify changes in children's health status and periodic risk signals during their growth stages. To address this issue, this invention generates a growth cycle factor by combining tanh and sine functions to simulate the complex and ever-changing health evolution trends during children's growth. The growth cycle modulation factor is combined element-wise with the original feature sequence, enabling the model to dynamically assign weights to different features according to actual growth patterns, highlighting signals with strong periodic correlations and significant changes, significantly suppressing the influence of occasional noise, and further utilizing a long short-term memory network to integrate health information from the current moment and multiple historical steps, effectively capturing long-term dependencies and improving the model's sensitivity and accuracy in identifying dynamic changes in children's health status.

[0036] Preferably, in step S4, constructing the growth cycle feature module includes the following steps:

[0037] S41. Generate the first... using tanh and the sine function. Growth cycle factor at time The mathematical model used to simulate the nonlinear physiological changes and periodic perturbations in children's growth process is as follows:

[0038] ;

[0039] In the formula, For the first Growth cycle factor at any given time The weight matrix is ​​a learnable matrix. For learnable bias terms, For learnable periodic frequencies, For learnable phase, To expand data dimensions through broadcasting mechanisms;

[0040] S42, Growth cycle modulation factor Characteristic sequences of children's health data By combining element-wise multiplication, the characteristics of strong periodic correlation and significant changes are highlighted. The specific mathematical model is as follows:

[0041] ;

[0042] In the formula, This is a physiological feature sequence after periodic modulation. This is element-wise multiplication;

[0043] S43, will Time to the The physiological characteristic sequence at time t is combined to form a sequence of length t. Time series data , Long Short-Term Memory (LSTM) networks are used to capture long-term dependencies. The specific mathematical model is as follows:

[0044] ;

[0045] In the formula, To utilize the time series data output by the Long Short-Term Memory network, ,in For the first The temporal feature sequence output by the Long Short-Term Memory Network contains historical information. It is a long short-term memory network.

[0046] Preferably, joint modeling using tanh and sinusoidal function modulation can simultaneously express the nonlinear physiological changes and periodic changes in the growth process, which helps to characterize the intrinsic evolutionary patterns in children's vital signs data, and the growth cycle modulation factor and original feature sequence. By combining elements one by one, the weights of feature expression can be dynamically adjusted, highlighting important signals with periodic correlations, suppressing noise interference, and enabling the model to obtain dynamic feature inputs that are more in line with physiological reality. Furthermore, by modeling time series features through a long short-term memory network, not only is the current state information captured, but historical time series dependencies are also preserved, enhancing the timeliness and accuracy of health status judgment.

[0047] Preferably, children's health status is often affected by both environmental and physiological changes, leading to sudden fluctuations or abnormal changes in health characteristics. Traditional feature processing methods often ignore these fluctuations, making the model slow to respond to abnormal signals and insensitive to short-term drastic changes. This invention measures the difference in physiological feature sequences between two consecutive time points to capture the dynamic fluctuation of health status in a timely manner. Furthermore, it uses nonlinear transformation to generate feature reconstruction parameters and adaptively adjusts the model's focus weight on current physiological feature information and historical time-series feature information according to the fluctuation intensity. While automatically sensing and adjusting health characteristics, it effectively improves the sensitivity and accuracy of response to abnormal fluctuations and sudden health risks.

[0048] Preferably, in step S5, constructing the health status fluctuation sensing module specifically includes the following steps:

[0049] S51, Calculate the first The difference between the time series feature sequence at a given time point and the time series feature sequence at the previous time point is used to capture the dynamic fluctuation value of health status. The specific mathematical model is as follows:

[0050] ;

[0051] In the formula, For the first Constant-time health status fluctuation value It is an L1 norm. For the first A sequence of physiological characteristics at any given moment;

[0052] S52. Using nonlinear transformation, feature reconstruction parameters are generated to adjust the attention weights on physiological feature sequences and time-series feature sequences containing historical information. The specific mathematical model is as follows:

[0053] ;

[0054] In the formula, For the first Feature reconstruction parameters at time step For learnable weights, For learnable bias terms;

[0055] S53, the first Temporal feature sequence of time step With physiological characteristic sequence The weighted fusion is performed based on feature reconstruction parameters, and the specific mathematical model is as follows:

[0056] ;

[0057] In the formula, This is the weighted fusion of health feature sequences.

[0058] Preferably, the L1 norm of the time-series feature sequence difference is used to accurately quantify the fluctuation range of the current health status, providing sensitivity support for abnormal changes or trend inflection points. By introducing feature reconstruction parameters based on the fluctuation range adaptively generated, the model's attention weight to the time-series and physiological feature sequences can be dynamically adjusted to achieve adaptive balance for sudden abnormalities or stable phases. Finally, the time-series and physiological feature sequences are weighted and fused according to the feature reconstruction parameters to improve the robustness and sensitivity of the model under changes in health status.

[0059] Preferably, in the child health monitoring method, the ultimate goal is to accurately and quantitatively evaluate the child's health status. To this end, the present invention flattens the health feature sequence into a feature vector, obtains the output vector through a combination of a fully connected network and the ReLU activation function, enhances the model's ability to express complex health patterns, and then further compresses the output vector and maps it into a single health monitoring value through another linear transformation, thereby realizing the quantification of the child's health monitoring value.

[0060] Preferably, in step S6, the health characteristic sequence is... Concatenate along the last dimension and flatten into a feature vector. The feature vector is sequentially input into a fully connected layer and ReLU activation to obtain the output vector. Then, a linear transformation is applied to obtain the health monitoring value. The specific mathematical model is as follows:

[0061] ;

[0062] ;

[0063] In the formula, For the output vector, , The weight matrix is ​​a learnable matrix. , For learnable bias terms, This is the final output of the child's health monitoring values.

[0064] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention assembles all data at each moment in children's vital signs data into a data sequence, utilizes lookup table mapping and linear transformation, and activation functions to achieve efficient extraction of high-dimensional feature sequences, combines adaptive weights and smoothing coefficient matrices, dynamically weights and smooths feature sequences, improving the model's adaptability to mutations, anomalies, and trend changes, introduces growth cycle factors to obtain physiological feature sequences, further utilizes long short-term memory networks to fully explore the historical evolution patterns of children's vital signs data to obtain temporal feature sequences, and obtains health feature sequences by weighted fusion of temporal and physiological feature sequences through feature reconstruction parameters, and finally, flattens the health feature sequences and inputs them into a fully connected neural network, activation functions, and linear transformations to obtain more comprehensive and accurate children's health monitoring values. Attached Figure Description

[0065] Figure 1 This is a flowchart illustrating the steps of a child health monitoring method based on vital sign data.

[0066] Figure 2 This is a diagram of the adaptive time-series weighting module.

[0067] Figure 3 This is a diagram of the growth cycle characteristics.

[0068] Figure 4 Diagram of the health status fluctuation sensing module.

[0069] Figure 5 A trend chart showing the change in training loss values ​​for children's health monitoring methods.

[0070] Figure 6 This is a graph showing the results of children's health monitoring. Detailed Implementation

[0071] This invention proposes a method for monitoring children's health based on vital sign data. This method organizes all data from each moment in a child's vital sign data into a data sequence. A lookup table is used to transform the data sequence into a high-dimensional implicit feature sequence. The feature sequence is further extracted through linear transformation and the ReLU activation function. An adaptive temporal weighting and smoothing adjustment mechanism is employed to dynamically adjust the feature sequence to obtain a weighted feature sequence. A growth cycle factor is introduced and multiplied element-wise with the weighted feature sequence to obtain a physiological feature sequence. A long short-term memory network is used to capture temporal features, obtaining a temporal feature sequence. The difference between the current and previous physiological feature sequences is calculated, and feature reconstruction parameters are introduced to adaptively weight and fuse the temporal and physiological features to form a health feature sequence. Finally, the health feature sequence is flattened and input into a fully connected neural network and activation function. After linear transformation, the child's health monitoring value is output, achieving comprehensive and accurate monitoring of children's health. The technical solutions in the embodiments of this invention will be described in detail and completely below, specifically including the following steps, such as... Figure 1 As shown.

[0072] S1. Collect vital signs data that affect children's health, preprocess them using the Z-score method, and construct a children's vital signs dataset.

[0073] Furthermore, in S1, data affecting children's health for 60 consecutive days are collected, including weight, height, body temperature, heart rate, blood oxygen saturation, ambient temperature, steps, and sleep duration. For each type of data, the Z-score method is used to detect outliers. Any data value with a Z-score greater than 3 or less than -3 is judged as an outlier. For the detected outliers, the mean replacement method is used to process them, that is, the outlier is replaced with the mean of the data three days before and after it. A children's vital signs dataset is constructed, and the dataset is divided into a training set and a validation set in a 7:3 ratio.

[0074] S2. All data at each moment in the child's vital signs dataset are combined into a data sequence. A lookup table is used to map the data sequence to a high-dimensional space to obtain an implicit feature sequence reflecting the child's health status. The feature sequence is obtained through linear transformation and ReLU activation function.

[0075] Furthermore, in S2, in the children's vital signs dataset, all data at each moment constitute the data sequence of the model input, for the first... A sequence of vital signs that constantly affect children's health , ,in For the first The collected values ​​of children's vital signs data, The number of data types related to children's vital signs. , the sequence of vital signs data The mapping to a richer feature sequence includes the following steps:

[0076] S21. Sequence of vital signs data By mapping a lookup table to a high-dimensional space, a high-dimensional implicit feature sequence is obtained, enabling the model to automatically extract potential health information. The specific mathematical model is as follows:

[0077] ;

[0078] In the formula, For the first Implicit feature sequences that constantly influence children's health data The dimension obtained by the lookup table mapping. For lookup table functions;

[0079] S22. Using linear transformation to transform implicit feature sequences Mapping to the physiological feature space and enhancing the model's ability to express nonlinear features through the ReLU activation function, the specific mathematical model is as follows:

[0080] ;

[0081] In the formula, For the transformed first Feature sequences that constantly influence children's health data A learnable weight matrix that maps implicit feature sequences to a physiological feature space. This is a learnable bias term.

[0082] S3. Construct an adaptive time-series weighting module. Based on the feature sequence and the weight of the previous time step, generate the weight of the current time step using nonlinear adjustment. At the same time, calculate the smoothing coefficient matrix, dynamically weight and smooth the feature sequence to obtain the weighted feature sequence.

[0083] Furthermore, in S3, an adaptive temporal weighting module is constructed, such as... Figure 2 As shown, the specific steps include:

[0084] S31, Comprehensive Section Weight of time and the first The information of the time-series features is combined with nonlinear adjustment to control the rate and magnitude of weight changes. The specific mathematical model is as follows:

[0085] ;

[0086] In the formula, For the first Weights of the time-series feature sequences For the first The weights of the feature sequence at each time step are initialized to a matrix of all 1s at the initial time step. The step size parameter is set to 0.05 during implementation to allow for learnable adjustment of the weight update magnitude. The weighted suppression parameter is a learnable parameter, and its initial value is set to 0.05 during implementation. The nonlinear modulation parameters are learnable, and their initial values ​​are set to 1 during implementation.

[0087] S32. Introduce a smoothing coefficient matrix. The size of the smoothing coefficient matrix is ​​automatically adjusted based on the change amplitude of the feature sequence at adjacent time points. The specific mathematical model is as follows:

[0088] ;

[0089] In the formula, For the first The smoothing coefficient matrix at time step, For the first Feature sequences that constantly influence children's health data As a learnable feature change-sensitive parameter, its initial value was set to 0.1 during implementation. This is for taking the absolute value;

[0090] S33, the first Time-smoothing coefficient matrix Weight With the Feature sequences of children's health data at any time Multiplying them yields a weighted feature sequence. The specific mathematical model is as follows:

[0091] ;

[0092] In the formula, For the first The weighted feature sequence at time step.

[0093] S4. Construct a growth cycle feature module, introduce a growth cycle factor, and multiply it element-by-element with the weighted feature sequence to obtain a physiological feature sequence. The physiological feature sequence forms time series data, and a long short-term memory network is used to obtain a time series feature sequence.

[0094] Furthermore, in step S4, a growth cycle feature module is constructed, such as... Figure 3 As shown, it includes the following steps:

[0095] S41. Generate the first... using tanh and the sine function. Growth cycle factor at time The specific mathematical model is as follows:

[0096] ;

[0097] In the formula, For the first Growth cycle factor at any given time The weight matrix is ​​a learnable matrix. For learnable bias terms, To determine the learnable periodic frequency, the initial value was set to 0.4 during implementation. To ensure a learnable phase, the initial value was set to 0.4 during implementation. To expand the data dimensions through a broadcast mechanism, the data dimensions in the function are expanded to [number] during implementation. dimension;

[0098] S42, Growth cycle modulation factor Characteristic sequences of children's health data Combining element-wise multiplication, the specific mathematical model is as follows:

[0099] ;

[0100] In the formula, This is a physiological feature sequence after periodic modulation. This is element-wise multiplication;

[0101] S43, will Time to the Combination of physiological characteristic sequences at time points Forming a length of Time series data , Long Short-Term Memory (LSTM) networks are used to capture long-term dependencies. The specific mathematical model is as follows:

[0102] ;

[0103] In the formula, To utilize the time series data output by the Long Short-Term Memory network, ,in For the first The temporal feature sequence output by the Long Short-Term Memory Network contains historical information. It is a long short-term memory network.

[0104] S5. Construct a health status fluctuation perception module. By calculating the difference between the physiological feature sequence at the current moment and the previous moment, obtain the health status fluctuation value. Introduce a feature reconstruction parameter. Based on this parameter, weightedly fuse the time series feature sequence and the physiological feature sequence to form a health feature sequence.

[0105] Furthermore, in S5, a health status fluctuation sensing module is constructed, such as... Figure 4 As shown, the specific steps include:

[0106] S51, Calculate the first The mathematical model for the difference between the temporal characteristic sequence at a given time and the physiological characteristic sequence at the previous time is as follows:

[0107] ;

[0108] In the formula, For the first Constant-time health status fluctuation value It is an L1 norm. For the first A sequence of physiological characteristics at any given moment;

[0109] S52. Using nonlinear transformation, feature reconstruction parameters are generated to adjust the attention weights on physiological feature sequences and time-series feature sequences containing historical information. The specific mathematical model is as follows:

[0110] ;

[0111] In the formula, For the first Feature reconstruction parameters at time step For learnable weights, For learnable bias terms;

[0112] S53, the first Temporal feature sequence of time step With physiological characteristic sequence The weighted fusion is performed based on feature reconstruction parameters, and the specific mathematical model is as follows:

[0113] ;

[0114] In the formula, This is the weighted fusion of health feature sequences.

[0115] S6. Flatten the health feature sequence into a feature vector, input it sequentially into a fully connected neural network and an activation function, and then obtain the final child health monitoring value through linear transformation.

[0116] Furthermore, in step S6, the health characteristic sequence is... Concatenate along the last dimension and flatten into a feature vector. The feature vector is sequentially input into a fully connected layer and ReLU activation to obtain the output vector. Then, a linear transformation is applied to obtain the health monitoring value. The specific mathematical model is as follows:

[0117] ;

[0118] ;

[0119] In the formula, For the output vector, , The weight matrix is ​​a learnable matrix. , For learnable bias terms, This is the final output of the child's health monitoring values.

[0120] Furthermore, the model was implemented using the Python 3.8 programming language and the PyTorch framework, running in a CUDA 11.3 environment. Training was performed using an NVIDIA 3090 24GB GPU, the loss function was mean squared error loss, and the optimizer was selected from [previous model name]. The initial learning rate is set to Batch size is The training rounds are .

[0121] Furthermore, the trend of the loss value during the training process of this method with the number of iterations is as follows: Figure 5 As shown, the vertical axis represents the training loss value, and the horizontal axis represents the training epochs. Figure 5 As can be seen, the training loss value shows a continuous decreasing trend with the increase of training rounds, indicating that the model is constantly optimizing parameters to minimize the loss function. The effectiveness of this method in monitoring children's health is as follows: Figure 6 As shown, the vertical axis represents children's health monitoring values ​​(out of 100), and the horizontal axis represents time (days). The gray dashed line in the figure represents the actual health assessment value of children, while the black solid line represents the children's health monitoring value generated by this method. The gray shaded area represents the difference range between the two. Figure 6 It can be seen that the trends of the monitored values ​​and the actual assessed values ​​are basically consistent, with a small error range, which proves the excellent performance of the proposed method in children's health monitoring tasks and verifies the accuracy and reliability of the method.

[0122] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. A method for monitoring children's health based on vital sign data, characterized in that, Includes the following steps: S1. Collect vital signs data that affect children's health, preprocess them using the Z-score method, and construct a children's vital signs dataset; S2. All data at each moment in the children's vital signs dataset are combined into a data sequence. A lookup table is used to map the data sequence to a high-dimensional space to obtain an implicit feature sequence reflecting the children's health status. The feature sequence is obtained through linear transformation and ReLU activation function. S3. Construct an adaptive temporal weighting module. Based on the feature sequence and the weights of the previous time step, generate the weights of the current time step using nonlinear adjustment. Simultaneously, introduce a smoothing coefficient matrix to dynamically weight and smooth the feature sequence, obtaining a weighted feature sequence. The adaptive temporal weighting module specifically includes the following steps: S31, Comprehensive Section Weight of time and the first The information of the time-series features, combined with the rate and magnitude of change of the nonlinear adjustment control weights, is used in the following mathematical model: ； In the formula, For the first Weights of the time-series feature sequences For the first Weights of the time-series feature sequences The step size parameter is used to adjust the magnitude of the learnable weight updates. For learnable weighted suppression parameters, These are learnable nonlinear modulation parameters. For the first Feature sequences that constantly influence children's health data; S32. Introduce a smoothing coefficient matrix. The size of the smoothing coefficient matrix is ​​automatically adjusted based on the change amplitude of the feature sequence at adjacent time points. The specific mathematical model is as follows: ； In the formula, For the first The smoothing coefficient matrix at time step, For the first Feature sequences that constantly influence children's health data For learnable feature changes sensitive parameters, This is for taking the absolute value; S33, the first Time-smoothing coefficient matrix With the Feature sequences of children's health data at any time Multiply, then multiply by the weight. Multiplying them together yields a weighted feature sequence. ; S4. Construct a growth cycle feature module, introduce a growth cycle factor, and multiply it element-wise with the weighted feature sequence to obtain a physiological feature sequence. The physiological feature sequence forms time series data, and a long short-term memory network is used to obtain the time series feature sequence. The growth cycle feature module specifically includes the following steps: S41. Generate the first... using tanh and the sine function. Growth cycle factor at time The specific mathematical model is as follows: ； In the formula, For the first Growth cycle factor at any given time The weight matrix is ​​a learnable matrix. For learnable bias terms, For learnable periodic frequencies, For learnable phase, To expand data dimensions through broadcasting mechanisms; S42, Growth cycle modulation factor Characteristic sequences of children's health data By combining elements-by-element multiplication, the physiological characteristic sequence after periodic modulation is obtained. ; S43, will Time to the The physiological characteristic sequence at time t is combined to form a sequence of length t. Time series data , By using Long Short-Term Memory (LSTM) networks to capture long-term dependencies, we can obtain the first... Temporal feature sequences containing historical information output by a Long Short-Term Memory Network ; S5. Construct a health status fluctuation sensing module. By calculating the difference between the physiological feature sequences at the current time and the previous time, the health status fluctuation value is obtained. A feature reconstruction parameter is introduced, and the time-series feature sequence and the physiological feature sequence are weighted and fused based on the parameter to form a health feature sequence. The health status fluctuation sensing module specifically includes the following steps: S51, Calculate the first The L1 norm of the difference between the temporal characteristic sequence at a given time and the physiological characteristic sequence at the previous time is used to obtain the health status fluctuation value. ; S52. Using nonlinear transformation, feature reconstruction parameters are generated. The specific mathematical model is as follows: ； In the formula, For the first Feature reconstruction parameters at time step For learnable weights, For learnable bias terms; S53, the first Temporal feature sequence of time step With physiological characteristic sequence The weighted fusion is performed based on feature reconstruction parameters, and the specific mathematical model is as follows: ； In the formula, The weighted fusion of health feature sequences; S6. Flatten the health feature sequence into a feature vector, input it sequentially into a fully connected neural network and an activation function, and then obtain the final child health monitoring value through linear transformation.

2. The method for monitoring children's health based on vital sign data according to claim 1, characterized in that, We collected vital signs data that affect children's health, including weight, height, body temperature, heart rate, blood oxygen saturation, ambient temperature, steps, and sleep duration. For each type of data, we used the Z-score method to detect outliers. Any data value with a Z-score greater than 3 or less than -3 was identified as an outlier. For the detected outliers, we used the mean replacement method, which replaced the outlier with the mean of the data three days before and after it, thus constructing a children's vital signs dataset.

3. The method for monitoring children's health based on vital sign data according to claim 2, characterized in that, In the children's vital signs dataset, all data at each time step constitute the data sequence input to the model. For the first... A sequence of vital signs that constantly affect children's health ,Will The mapping to a richer feature sequence includes the following steps: S21. Sequence of vital signs data By mapping to a high-dimensional space using a lookup table, high-dimensional implicit feature sequences can be obtained. ; S22. Using linear transformation to transform implicit feature sequences Mapped to the physiological feature space and subjected to nonlinear processing using the ReLU activation function, the th... Feature sequences that constantly affect children's health data .

4. The method for monitoring children's health based on vital sign data according to claim 3, characterized in that, health characteristic sequence Concatenate along the last dimension and flatten into a feature vector. The feature vector is sequentially input into a fully connected layer and ReLU activation to obtain the output vector, which is then transformed linearly to obtain the child's health monitoring value. .

Images