A health monitoring system based on the Internet of Things

Abstract

The present invention relates to the field of remote monitoring technology, specifically a health monitoring system based on the Internet of Things, the system comprising: a morphological feature capture module separates a QRS complex wave from an electrocardiogram signal, and measures morphological features including shape, width and height, applies open and close operations and morphological gradients to the morphological features, and extracts waveform data. In the present invention, by integrating long short-term memory networks and conditional variational autoencoders, the monitoring accuracy and dynamic response capability of cardiopulmonary function are improved, the data processing logic is made more refined, and the slight changes in the patient's physiological state can be accurately captured through continuous data comparison and dynamic reconstruction technology, and potential health risks can be identified at an early stage, providing a more personalized health status assessment. This integrated application is excellent in improving the practicality and predictive value of data, so that the monitoring results are not limited to simple data records, but can provide decision support.

Description

Technical Field

[0001] The present invention relates to the field of remote monitoring technology, and in particular to a health monitoring system based on the Internet of Things. Background Art

[0002] Remote monitoring technology involves the use of various sensors and devices to collect data over long distances and transmit this data to a central system through a communication network for analysis and processing. This technology is widely used in many fields such as healthcare, environmental monitoring, and industrial automation. In the medical field, remote monitoring can achieve real-time tracking of patients' physiological parameters, such as heart rate, blood pressure, etc., allowing doctors to diagnose and monitor patients without being in the hospital. The application of this technology has greatly improved the accessibility and efficiency of medical services, especially in remote areas and when there is a high demand for home care.

[0003] Among them, health monitoring systems refer to technical systems that use various sensors to measure and track individual health conditions. The uses of these systems include, but are not limited to, continuous monitoring of vital signs, diagnosis of health problems, prevention of diseases, and optimization of treatment plans. By collecting and analyzing real-time data from patients, health monitoring systems can help medical professionals respond quickly, improve patient treatment outcomes, and support telemedicine services, allowing medical resources to serve a larger population more widely. This system is particularly suitable for chronic patients and the elderly, providing them with an effective tool for continuous monitoring of their health.

[0004] Existing remote monitoring technologies mainly focus on the collection and transmission of basic data, and lack the ability to conduct in-depth analysis and real-time prediction of data. When processing complex physiological signals, such as electrocardiogram waveform analysis, conventional methods usually fail to effectively identify or predict potential abnormal patterns, limiting their application in actual medical scenarios. Especially when long-term health monitoring based on individual differences is required, existing systems often fail to take into account the dynamic changes of individuals, resulting in a lack of personalization and accuracy in health assessments. The limitations of this technology not only slow down the timeliness of medical decision-making, but may also increase the risk of misdiagnosis in emergency medical situations, thereby affecting the treatment effect and safety of patients. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a health monitoring system based on the Internet of Things.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: A health monitoring system based on the Internet of Things includes:

[0007] The morphological feature capture module separates the QRS complex wave from the ECG signal and measures the morphological features including shape, width and height, applies opening and closing operations and morphological gradients to the morphological features, extracts waveform data, and obtains key morphological features;

[0008] The ECG pattern analysis module is based on the key morphological features, serializes the ECG data and inputs it into the long short-term memory network, calculates the sample entropy and approximate entropy of the ECG signal cycle multiple times, aggregates the Lyapunov exponent of each ECG cycle, and analyzes it through phase space reconstruction technology to obtain the ECG behavior pattern;

[0009] The dynamic analysis module of lung function inputs the patient's age, medical history and lung function data into the conditional variational autoencoder for feature extraction, inputs the extracted features into the bidirectional long short-term memory network, analyzes the dynamic changes of time dependence, and reconstructs and compares the current lung function data with the historical data to obtain the lung function variation characteristics;

[0010] The data joint application module analyzes and compares the electrocardiographic behavior pattern and the lung function variation characteristics, analyzes the correlation and time series changes of the two data, evaluates the health status, and obtains the health assessment result.

[0011] As a further solution of the present invention, the step of acquiring the key morphological features is specifically as follows:

[0012] The Pan-Tompkins algorithm is used to locate the peak of the R wave in the ECG signal using the formula:

[0013] R peak =max(s(t))·k

[0014] Generate R wave peak position R peak ;

[0015] Among them, R peak represents the peak position of the R wave, s(t) represents the ECG signal, t represents time, and k represents the calibration coefficient;

[0016] Using the R wave peak position R peak , measure the width and height of the QRS complex wave, using the formula:

[0017]

[0018] and

[0019]

[0020] Generate QRS complex width and height QRS_dimensions;

[0021] Where W represents width, H represents height, t(Q) and t(S) represent the time positions of Q wave and S wave respectively, c represents the time scaling factor, and a represents the amplitude magnification factor;

[0022] The QRS complex width and height QRS_dimensions are applied to perform morphological operations, identify waveforms and extract features, using the formula:

[0023]

[0024] Obtain the key morphological feature G;

[0025] in, represents the time derivative of the ECG signal s(t), open(s(t)) and close(s(t)) represent the opening operation and closing operation respectively, and d represents the normalization coefficient after the operation.

[0026] As a further solution of the present invention, the steps of obtaining the sample entropy and approximate entropy of the ECG signal period are specifically as follows:

[0027] Based on the key morphological features of ECG data, it is input into the long short-term memory network for processing, and the data serialization process is optimized using the formula:

[0028]

[0029] Generate processed serialized ECG data X;

[0030] Among them, LSTM stands for long short-term memory network processing function, α, β, γ are network adjustment parameters, and δ is the adjustment factor;

[0031] The serialized ECG data is applied to calculate sample entropy and approximate entropy at the same time, and the accuracy of data analysis is enhanced through mathematical models, using the formula:

[0032]

[0033] and

[0034]

[0035] Get the sample entropy and approximate entropy of the ECG cycle;

[0036] Among them, S represents sample entropy, A represents approximate entropy, and p i represents the probability of the sequence, n is the total number of subsequences, is the fitness calculation function, m is the template length, r is the fitness threshold, k is the adjustment weight, and λ and μ are normalization coefficients.

[0037] As a further solution of the present invention, the step of acquiring the electrocardiographic behavior pattern is specifically as follows:

[0038] Aggregate the sample entropy and approximate entropy of the ECG cycle to evaluate the behavioral characteristics of the overall ECG signal using the formula:

[0039]

[0040] Calculate the average value of the sum of sample entropy of the ECG signal period and the square root of the sum of squared approximate entropy to generate an entropy feature evaluation E of the ECG signal;

[0041] Among them, S i is the sample entropy of the ith period, A j is the approximate entropy of the jth period, n is the total number of sample entropies, and m is the total number of approximate entropies;

[0042] The entropy characteristic evaluation of the ECG signal is input into the phase space reconstruction technology to analyze the ECG behavior pattern using the formula:

[0043]

[0044] Generate electrocardiographic behavior pattern M;

[0045] Among them, cos is the cosine function, τ k represents the time delay coefficient, E k represents the entropy feature of the corresponding time point, and p is the total number of time points.

[0046] As a further solution of the present invention, the feature extraction step is specifically as follows:

[0047] The patient's age, medical history, and lung function data are input into the encoder part of the conditional variational autoencoder, and the data is standardized using the formula:

[0048]

[0049] Generate standardized input data z norm ;

[0050] Among them, x represents the original input data, μ x represents the mean value of the data, σ x represents the standard deviation of the data, ∈ represents a constant to avoid division by zero error;

[0051] A weighted and biased nonlinear transformation is applied to the normalized input data using the formula:

[0052] z=tanh(W·z norm +b+α·sin(β·z norm ))

[0053] Calculate the latent variables and generate the latent space mapping result z;

[0054] Where W represents the weight of the encoder, b represents the bias of the encoder, and α and β are adjustment parameters;

[0055] The latent space mapping result is used to decode in the decoder, using the formula:

[0056] x ′ =V·z / (1+|c·z|)

[0057] The introduction of fractions and absolute values ​​increases the nonlinear characteristics of the model, the reconstructed data is compared with the original data, the reconstruction error is calculated, and the reconstructed data x is generated. ′ ;

[0058] Among them, V represents the weight of the decoder, and c represents the bias of the decoder;

[0059] The reconstructed data is normalized using the formula:

[0060]

[0061] Perform feature extraction F;

[0062] Among them, min(x ′ ) represents the minimum value of the reconstructed data, max(x ′ ) represents the maximum value of the reconstructed data, and δ represents a constant to avoid the denominator being zero.

[0063] As a further embodiment of the present invention, the step of acquiring the lung function variation characteristics is specifically as follows:

[0064] The feature extraction results are integrated with the historical lung function data set, and a composite calculation formula is used:

[0065]

[0066] Generate integrated data set D;

[0067] Among them, H represents the feature set of historical lung function data, ω and ω′ represent weight coefficients, and λ represents the balanced nonlinear adjustment coefficient;

[0068] The bidirectional long short-term memory network is applied to process the integrated data set D and analyze the dynamic changes of time dependence using the formula:

[0069] T=BiLSTM(D,α,β,γ)

[0070] Generate time-dependent feature analysis result T;

[0071] Among them, BiLSTM stands for bidirectional long short-term memory network, α, β, and γ are network adjustment parameters;

[0072] The time-dependent features were compared with the historical lung function data set and the variation features were calculated using the formula:

[0073]

[0074] Generate lung function variation signature V;

[0075] Where H represents the historical lung function dataset and κ is the coefficient for adjusting the sensitivity of the difference.

[0076] As a further solution of the present invention, the steps of obtaining the health assessment results are specifically as follows:

[0077] From the time series data of the ECG behavior pattern and the lung function variability characteristics, the formula is used:

[0078]

[0079] Calculate and generate the normalized correlation coefficient R;

[0080] Among them, M i Represents the value of the i-th ECG behavior pattern, V i represents the value of the ith lung function variation feature, σ M , σ V represents the standard deviation of the ECG behavior pattern and lung function variability characteristics, ∈ represents a small constant to avoid division by zero;

[0081] The normalized correlation coefficient is subjected to time window analysis using the formula:

[0082]

[0083] Calculate and generate weighted cumulative correlation P;

[0084] Among them, R(t) represents the normalized correlation coefficient of time, δ represents the attenuation factor, and T represents the total observation time;

[0085] The health assessment is performed based on the weighted cumulative association, using the formula:

[0086]

[0087] Generate health assessment result U;

[0088] Where e represents the base of the natural logarithm and k represents a constant that adjusts the response rate.

[0089] Compared with the prior art, the advantages and positive effects of the present invention are:

[0090] In the present invention, by integrating long short-term memory networks and conditional variational autoencoders, the monitoring accuracy and dynamic response capability of cardiopulmonary function are improved. The data processing logic is made more refined, and it is possible to accurately capture subtle changes in the patient's physiological state through continuous data comparison and dynamic reconstruction technology. Using this method, the health monitoring system can identify potential health risks at an early stage and provide a more personalized health status assessment. This integrated application excels in improving the practicality and predictive value of data, so that the monitoring results are not limited to simple data records, but can provide decision support, thereby significantly optimizing the medical response speed and the formulation of treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 is a system flow chart of the present invention;

[0092] Figure 2 Flow chart of the steps for obtaining the key morphological features of the present invention;

[0093] Figure 3 A flow chart of the steps for obtaining sample entropy and approximate entropy of an electrocardiogram signal cycle according to the present invention;

[0094] Figure 4 A flowchart of the steps for obtaining the electrocardiographic behavior pattern of the present invention;

[0095] Figure 5 It is a flow chart of the steps of feature extraction of the present invention;

[0096] Figure 6 is a flow chart of the steps for obtaining the lung function variation characteristics of the present invention;

[0097] Figure 7 The figure is a flow chart of the steps for obtaining the health assessment results of the present invention. DETAILED DESCRIPTION

[0098] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0099] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0100] See also Figure 1 , a health monitoring system based on the Internet of Things includes:

[0101] The morphological feature capture module separates the QRS complex wave from the ECG signal and measures the morphological features including shape, width and height, applies opening and closing operations and morphological gradients to the morphological features, extracts waveform data, and obtains key morphological features;

[0102] The ECG pattern analysis module is based on key morphological features, serializes ECG data and inputs it into the long short-term memory network, calculates the sample entropy and approximate entropy of the ECG signal cycle multiple times, aggregates the Lyapunov exponent of each ECG cycle, and analyzes it through phase space reconstruction technology to obtain the ECG behavior pattern;

[0103] The dynamic analysis module of lung function inputs the patient's age, medical history and lung function data into the conditional variational autoencoder for feature extraction, inputs the extracted features into the bidirectional long short-term memory network, analyzes the dynamic changes of time dependence, and reconstructs and compares the current lung function data with the historical data to obtain the lung function variation characteristics;

[0104] The data joint application module analyzes and compares the ECG behavior patterns and lung function variation characteristics, analyzes the correlation and time series changes of the two data, evaluates the health status, and obtains the health assessment results.

[0105] The key morphological features are specifically the waveform peak, waveform width, and waveform height. The ECG behavior patterns include the cycle entropy value, the cycle Lyapunov exponent, and the time series reconstruction results. The pulmonary function variation characteristics specifically refer to the time-dependent changes, data reconstruction differences, and feature extraction results. The health assessment results include cardiopulmonary correlation analysis, health status time series, and comprehensive evaluation indicators.

[0106] See also Figure 2 , the steps to obtain key morphological features are as follows:

[0107] The Pan-Tompkins algorithm is used to locate the peak of the R wave in the ECG signal using the formula:

[0108] R peak =max(s(t))·k

[0109] Generate R wave peak position R peak ;

[0110] Among them, R peak represents the peak position of the R wave, s(t) represents the ECG signal, t represents time, and k represents the calibration coefficient;

[0111] Use the R wave peak position R peak , measure the width and height of the QRS complex wave, using the formula:

[0112]

[0113] and

[0114]

[0115] Generate QRS complex width and height QRS_dimensions;

[0116] Where W represents width, H represents height, t(Q) and t(S) represent the time positions of Q wave and S wave respectively, c represents the time scaling factor, and a represents the amplitude magnification factor;

[0117] Apply QRS complex width and height QRS_dimensions to perform morphological operations, identify waveforms and extract features using the formula:

[0118]

[0119] Obtain the key morphological feature G;

[0120] in, represents the time derivative of the ECG signal s(t), open(s(t)) and close(s(t)) represent the opening operation and closing operation respectively, and d represents the normalization coefficient after the operation.

[0121] formula:

[0122] R peak =max(s(t))·k

[0123] in:

[0124] s(t): ECG signal, which is a function representing the voltage changing over time.

[0125] t: time variable.

[0126] k: Calibration factor used to adjust the sensitivity of peak detection.

[0127] Calculation process and examples:

[0128] Assume that the expression of the ECG signal s(t) in a certain time period is s(t)=1.5sin(60πt), where the unit of t is second.

[0129] We need to find the maximum value of s(t). For a sine wave, the maximum value is the amplitude, which is 1.5.

[0130] Set the calibration factor k=1.2 (this factor is set according to the equipment and measurement accuracy, between 1 and 2).

[0131] Use the formula to calculate the R wave peak position:

[0132] R peak =1.5×1.2=1.8

[0133] Wherein, 1.8 represents the adjusted R wave peak voltage in millivolts.

[0134] formula:

[0135]

[0136] and

[0137]

[0138] in:

[0139] t(Q), t(S): represent the time position of Q wave and S wave respectively.

[0140] c: Time scaling factor, used to adjust the width of the measurement.

[0141] a: Amplitude magnification factor, used to adjust the height calculation.

[0142] Calculation process and examples:

[0143] Assume t(Q)=0.2 seconds, t(S)=0.35 seconds, and c=2.

[0144] The calculation of QRS width is:

[0145]

[0146] Here, 0.075 seconds represents the duration of the QRS complex.

[0147] Assume max(s(t))=1.5 mV, min(s(t))=-1.5 mV, and a=4.

[0148] The calculation of QRS height is:

[0149]

[0150] Here, 3 mV represents the height of the adjusted QRS complex.

[0151] formula:

[0152]

[0153] in:

[0154] The time derivative of the ECG signal.

[0155] open(s(t)), close(s(t)): opening and closing operations, used for morphologically processed signals.

[0156] d: Normalization coefficient.

[0157] Calculation process and examples:

[0158] Assume that the derivative of the ECG signal is approximately At t=0, this value is 90π.

[0159] Assuming that the opening and closing operations do not change the maximum amplitude of the signal, let open(s(t))+close(s(t))=3.

[0160] Set d=1.5.

[0161] The morphological gradient is calculated as:

[0162]

[0163] Here, 565.5 represents the measure of the morphological gradient in millivolts per second.

[0164] See also Figure 3 , the steps for obtaining the sample entropy and approximate entropy of the ECG signal cycle are as follows:

[0165] Based on the key morphological features of ECG data, it is input into the long short-term memory network for processing, and the data serialization process is optimized using the formula:

[0166]

[0167] Generate processed serialized ECG data X;

[0168] Among them, LSTM stands for long short-term memory network processing function, α, β, γ are network adjustment parameters, and δ is the adjustment factor;

[0169] Apply serialized ECG data, calculate sample entropy and approximate entropy at the same time, and enhance the accuracy of data analysis through mathematical models. The formula is:

[0170]

[0171] and

[0172]

[0173] Get the sample entropy and approximate entropy of the ECG cycle;

[0174] Among them, S represents sample entropy, A represents approximate entropy, and p i represents the probability of the sequence, n is the total number of subsequences, is the fitness calculation function, m is the template length, r is the fitness threshold, k is the adjustment weight, and λ and μ are normalization coefficients.

[0175] formula:

[0176]

[0177] Detailed explanation:

[0178] LSTM(G,α,β,γ): This is a long short-term memory network used to process the input ECG signal feature G, where α, β, and γ are the parameters of the network's learning rate, forget gate, and output gate. It is assumed that the parameters control the network's learning process and memory retention.

[0179] δ: Used as a normalization factor to ensure that the output X is within a matching numerical range and avoid the gradient vanishing or exploding problem.

[0180] Assume parameter settings:

[0181] α=0.01, β=0.8, γ=0.5

[0182] δ=0.1

[0183] G = 0.75 (indicates the standardized value of a certain ECG characteristic)

[0184] Calculation process:

[0185] Input G to the LSTM network.

[0186] The LSTM network calculates the output based on the provided parameters, which can be simplified to LSTM(G,α,β,γ)=αG+βG 2 -γ

[0187] Substituting the numbers, LSTM(0.75,0.01,0.8,0.5)=0.01×0.75+0.8×0.75 2 -0.5 = 0.0075 + 0.45 - 0.5 = -0.0425.

[0188] Apply a normalization factor δ to the LSTM output,

[0189] Example results:

[0190] The result X=-0.425 represents the characteristic strength of the normalized ECG signal, and a negative value represents an abnormal ECG activity or an activity below the average level.

[0191] formula:

[0192]

[0193] and

[0194]

[0195] Detailed explanation:

[0196] S: represents sample entropy, which is used to measure the complexity of ECG signals, where p i is the probability of occurrence of the sequence.

[0197] A: represents the approximate entropy, which measures the difficulty of predicting the signal. It is the ratio of similar template matches, which is used to calculate the regularity of the data.

[0198] λ, μ, k: Normalization coefficients and adjustment weights used to control the accuracy and sensitivity of the calculation.

[0199] Assume parameter settings:

[0200] n=3, m=2, k=2

[0201] p1=0.5, p2=0.3, p3=0.2

[0202]

[0203] λ=2,μ=0.5,r=0.05

[0204] Calculation process:

[0205] Calculate sample entropy:

[0206] Calculate the approximate entropy:

[0207] Plug in numbers and calculate the result.

[0208] Example results:

[0209] The values ​​of sample entropy and approximate entropy are used to describe the complexity and prediction difficulty of ECG signals. Higher sample entropy or approximate entropy values ​​indicate increased irregularity and unpredictability of ECG signals, which helps identify abnormal ECG patterns or disease states.

[0210] See also Figure 4 , the specific steps for obtaining the ECG behavior pattern are:

[0211] Aggregate the sample entropy and approximate entropy of the ECG cycle to evaluate the behavioral characteristics of the overall ECG signal using the formula:

[0212]

[0213] Calculate the average value of the sum of sample entropy of the ECG signal period and the square root of the sum of squared approximate entropy to generate an entropy feature evaluation E of the ECG signal;

[0214] Among them, S i is the sample entropy of the ith period, A j is the approximate entropy of the jth period, n is the total number of sample entropies, and m is the total number of approximate entropies;

[0215] The entropy characteristic evaluation of the ECG signal is input into the phase space reconstruction technique to analyze the ECG behavior pattern using the formula:

[0216]

[0217] Generate electrocardiographic behavior pattern M;

[0218] Among them, cos is the cosine function, τ k represents the time delay coefficient, E k represents the entropy feature of the corresponding time point, and p is the total number of time points.

[0219] formula:

[0220]

[0221] Parameter Description:

[0222] S i : Sample entropy of the ith period.

[0223] A j : Approximate entropy of the j-th period.

[0224] n: The total number of sample entropy.

[0225] m: The total amount of approximate entropy.

[0226] Assumptions and examples:

[0227] Assume there are 3 sample entropy values ​​and 3 approximate entropy values:

[0228] S = [0.5, 0.6, 0.55]

[0229] A=[0.4,0.45,0.35]

[0230] n=3

[0231] m=3

[0232] Calculation process:

[0233] Calculate the sum of sample entropy:

[0234] Compute the sum of squares of the approximate entropy:

[0235] Compute the square root of the sum of squared approximate entropy:

[0236] This evaluates the entire expression:

[0237] Result analysis:

[0238] The calculated E value is 0.391, which is a quantitative indicator of the overall entropy characteristics of the ECG signal. Lower values ​​indicate less irregularity or complexity.

[0239] formula:

[0240]

[0241] Parameter Description:

[0242] E k : Entropy feature at the kth time point.

[0243] τ k : The time delay coefficient at the kth time point.

[0244] p: total number of time points for reference.

[0245] Assumptions and examples:

[0246] Assume that there are 3 time points, the entropy feature E and time delay coefficient of each point are as follows:

[0247] E = [0.39, 0.42, 0.37]

[0248] τ=[1,2,1.5]

[0249] p=3

[0250] Calculation process:

[0251] Calculate the function value at each time point:

[0252] cos(1×0.39)≈cos(0.39)≈0.92

[0253] cos(2×0.42)≈cos(0.84)≈0.67

[0254] cos(1.5×0.37)≈cos(0.555)≈0.85

[0255] Calculate the average:

[0256] Result analysis:

[0257] The calculated M value is 0.813, which represents a measure of the average ECG behavior pattern extracted from the entropy characteristics of the ECG signal. A higher M value indicates that the ECG signal shows certain periodicity or regularity characteristics.

[0258] See also Figure 5 , the steps of feature extraction are as follows:

[0259] The patient's age, medical history, and lung function data are input into the encoder part of the conditional variational autoencoder, and the data is standardized using the formula:

[0260]

[0261] Generate standardized input data z norm ;

[0262] Among them, x represents the original input data, μ x represents the mean value of the data, σ x represents the standard deviation of the data, ∈ represents a constant to avoid division by zero error;

[0263] Apply a weighted and biased nonlinear transformation to the normalized input data using the formula:

[0264] z=tanh(W·z norm +b+α·sin(β·z norm ))

[0265] Calculate the latent variables and generate the latent space mapping result z;

[0266] Where W represents the weight of the encoder, b represents the bias of the encoder, and α and β are adjustment parameters;

[0267] Use the latent space mapping result to decode in the decoder, using the formula:

[0268] x ′ =V·z / (1+|c·z|)

[0269] The introduction of fractions and absolute values ​​increases the nonlinear characteristics of the model, the reconstructed data is compared with the original data, the reconstruction error is calculated, and the reconstructed data x is generated. ′ ;

[0270] Among them, V represents the weight of the decoder, and c represents the bias of the decoder;

[0271] Normalize the reconstructed data using the formula:

[0272]

[0273] Perform feature extraction F;

[0274] Among them, min(x ′ ) represents the minimum value of the reconstructed data, max(x ′ ) represents the maximum value of the reconstructed data, and δ represents a constant to avoid the denominator being zero.

[0275] Formula and derivation

[0276]

[0277] x: Assume that the original data is the patient's lung function index, take 5 sample points: x = [200, 220, 210, 230, 240]

[0278] μ x : The average value of the data. Calculation formula:

[0279] σ x : Standard deviation of the data. Calculation formula:

[0280]

[0281] ∈: A small constant to avoid division by zero error, ∈=0.0001

[0282] By substituting the values, we get:

[0283]

[0284] z norm =[-1.42,0,-0.71,0.71,1.42]

[0285] Result interpretation

[0286] The normalized result z norm , providing a standardized form, eliminating the magnitude effect of the original data and making the data comparable.

[0287] Formula and derivation

[0288] z=tanh(W·z norm +b+α·sin(β·z norm ))

[0289] Assumptions:

[0290] W=1.5, b=0.1, α=0.5, β=2, the value is determined by network training.

[0291] sin(β·z norm ): sin(2·[-1.42,0,-0.71,0.71,1.42])=sin([-2.84,0,-1.42,1.42,2.84])

[0292] =[-0.334,0,-0.991,0.991,0.334]

[0293] By substituting the values, we get:

[0294] z=tanh(1.5·[-1.42,0,-0.71,0.71,1.42]+0.1+0.5

[0295] [-0.334,0,-0.991,0.991,0.334])

[0296] z=tanh([-2.13,0.1,-1.36,1.36,2.13])

[0297] z=[-0.97,0.1,-0.88,0.88,0.97]

[0298] Result interpretation

[0299] The data is iteratively compressed into the latent space through nonlinear transformation, where z is the latent variable that captures the key features of the input data.

[0300] Formula and derivation

[0301] x ′ =V·z / (1+|c·z|)

[0302] Assumptions:

[0303] V=1.0, c=0.5, the value is determined by network training.

[0304] |c·z|:|0.5·[-0.97,0.1,-0.88,0.88,0.97]|=[0.485,0.05,0.44,0.44,0.485]

[0305] By substituting the values, we get:

[0306] x ′ =1.0 · [-0.97, 0.1, -0.88, 0.88, 0.97] / (1 + [0.485, 0.05, 0.44, 0.44, 0.485])

[0307] x′ =[-0.97,0.1,-0.88,0.88,0.97] / [1.485,1.05,1.44,1.44,1.485]

[0308] x ′ =[-0.65,0.095,-0.61,0.61,0.65]

[0309] Result interpretation

[0310] This reconstructed data x ′ This represents an attempt to decode from the latent space back to the original data space, which, while not exactly identical to the original data, should be close to the original data structure.

[0311] Formula and derivation

[0312]

[0313] Assumptions:

[0314] δ=0.001

[0315] By substituting the values, we get:

[0316]

[0317] F=[0,0.574,0.024,0.926,1]

[0318] Result interpretation

[0319] The normalized result, F, provides an expression from decoded data to feature extraction, where 1 represents the maximum normalized value and 0 represents the minimum normalized value, revealing the variability and relative position of the data.

[0320] See also Figure 6 , the specific steps for obtaining the lung function variation characteristics are:

[0321] Integrate the feature extraction results with the historical lung function data set and use a composite calculation formula:

[0322]

[0323] Generate integrated data set D;

[0324] Among them, H represents the feature set of historical lung function data, ω, ω ′ represents the weight coefficient, and λ represents the balance nonlinear adjustment coefficient;

[0325] The bidirectional long short-term memory network is used to process the integrated data set D and analyze the dynamic changes of time dependence. The formula is:

[0326] T=BiLSTM(D,α,β,γ)

[0327] Generate time-dependent feature analysis result T;

[0328] Among them, BiLSTM stands for bidirectional long short-term memory network, α, β, and γ are network adjustment parameters;

[0329] Compare the time-dependent features with the historical lung function data set and calculate the variation features using the formula:

[0330]

[0331] Generate lung function variation signature V;

[0332] Where H represents the historical lung function dataset and κ is the coefficient for adjusting the sensitivity of the difference.

[0333] Formula analysis and derivation

[0334]

[0335] in:

[0336] ω、ω ′ : These two parameters are weight coefficients, which are used to adjust the influence of current data F and historical data H in the integration process.

[0337] λ: This is a nonlinear coefficient that adjusts the balance and determines how the influence of the difference between F and H affects the nonlinear adjustment of the integration result.

[0338] F: The value of the current feature extraction result.

[0339] H: The numerical value of the feature set of historical lung function data.

[0340] D: The integrated feature dataset.

[0341] Calculation example

[0342] Assumptions:

[0343] F=120, current lung function characteristic score.

[0344] H = 100, historical average lung function characteristic score.

[0345] ω=0.6, the weight of the current data.

[0346] ω ′ =0.4, the weight of historical data.

[0347] λ=0.1, nonlinear adjustment coefficient.

[0348] Calculation process:

[0349] Calculate the weighted sum: ωF+ω ′ H=0.6×120+0.4×100=72+40=112

[0350] Calculate the nonlinear adjustment term: e ；λ(F；H) =e ；0.1×(120；100) =e ；2 ≈0.135

[0351] Calculate the integrated value:

[0352] Among them, D≈98.68 represents the adjusted lung function score after referring to current and historical data, indicating the integrated value of lung function characteristics under given weights and adjustment parameters.

[0353] Formula analysis and derivation

[0354] T=BiLSTM(D,α,β,γ)

[0355] in:

[0356] D: Integrate feature datasets.

[0357] α, β, γ: Tuning parameters of the network that control multiple aspects of the network, including the behavior of the forget gate, input gate, and output gate.

[0358] T: time-dependent features, representing the time series features extracted from the data.

[0359] Calculation example

[0360] Assumptions:

[0361] D=98.68.

[0362] α=0.5,β=0.3,γ=0.2, network parameters.

[0363] Due to the neural network structure involved in the calculation of the BiLSTM network, the internal calculation of the network is not shown, but it can be assumed that the network output is T = 95, which means that under the given network parameters and input, the extracted time-dependent feature is 95, which represents the time-dependent dynamic characteristics of the current lung function.

[0364] Formula analysis and derivation

[0365]

[0366] in:

[0367] T: time-dependent features.

[0368] H: Eigenvalue of the historical lung function dataset.

[0369] κ: A coefficient that adjusts the sensitivity of differences, enhancing or weakening the performance of differences.

[0370] V: represents the calculated lung function variation characteristics.

[0371] Calculation example

[0372] Assumptions:

[0373] T=95, time-dependent characteristics.

[0374] H=100, historical lung function characteristic value.

[0375] κ=1. To simplify the calculation, the parameter is assumed to be 1.

[0376] Calculation process:

[0377] Calculate the cube of the difference: T 3 -H 3 =95 3 -100 3 =857375-1000000=-142625

[0378] Take the absolute value and apply the adjustment factor: |κ×(-142625)|=|1×-142625|=142625

[0379] Apply the square root to get the mutation characteristics:

[0380] Among them, V≈377.66 represents the variability between current and historical lung function data under given parameters. The value of 377.66 is regarded as an indicator to measure the difference between the two, and the larger the value, the more obvious the difference.

[0381] See also Figure 7 , the specific steps for obtaining health assessment results are:

[0382] From the time series data of ECG behavior patterns and lung function variability characteristics, the formula is used:

[0383]

[0384] Calculate and generate the normalized correlation coefficient R;

[0385] Among them, M i Represents the value of the i-th ECG behavior pattern, V i represents the value of the ith lung function variation feature, σ M , σ V represents the standard deviation of the ECG behavior pattern and lung function variability characteristics, ∈ represents a small constant to avoid division by zero;

[0386] The normalized correlation coefficient is analyzed in time window using the formula:

[0387]

[0388] Calculate and generate weighted cumulative correlation P;

[0389] Among them, R(t) represents the normalized correlation coefficient of time, δ represents the attenuation factor, and T represents the total observation time;

[0390] Health assessment is performed based on weighted cumulative association using the formula:

[0391]

[0392] Generate health assessment result U;

[0393] Where e represents the base of the natural logarithm and k represents a constant that adjusts the response rate.

[0394] formula:

[0395]

[0396] Analysis and examples:

[0397] Assume that there are three groups of observed data points of ECG behavior pattern M and lung function variation characteristics V, which are:

[0398] M = [1.2, 0.5, 0.9]

[0399] V = [2.0, 1.5, 1.8]

[0400] Assume σ M (Standard deviation of ECG behavior pattern) is 0.35, σ V (standard deviation of the lung function variability characteristic) is 0.45, and ∈ (small constant to avoid zero) is 0.01.

[0401] Calculation steps:

[0402] Calculate each pair of M i and V i The product of:

[0403] M1×V1=1.2×2.0=2.4

[0404] M2×V2=0.5×1.5=0.75

[0405] M3×V3=0.9×1.8=1.62

[0406] Apply the formula for the normalized correlation coefficient:

[0407]

[0408] Sum the values:

[0409] R = 14.24 + 4.44 + 9.60 = 28.28

[0410] Result analysis:

[0411] The value of the normalized correlation coefficient R (28.28) indicates the overall correlation between the ECG behavior pattern and the pulmonary function variability characteristics, with higher values ​​indicating stronger associations.

[0412] formula:

[0413]

[0414] Analysis and examples:

[0415] Assume that the total observation time T is 3 hours and the attenuation factor δ is 0.1.

[0416] Calculation steps:

[0417] Assume R(t) is the previously calculated R, and evaluate R(t) every hour:

[0418] R(1)=R(2)=R(3)=28.28

[0419] The reference attenuation is calculated by weighting:

[0420] R(1)×e ；0.1×1 ≈28.28×0.905=25.61

[0421] R(2)×e ；0.1×2 ≈28.28×0.819=23.17

[0422] R(3)×e ；0.1×3 ≈28.28×0.741=20.95

[0423] Sum and divide by T:

[0424]

[0425] Result analysis:

[0426] The value of weighted cumulative correlation P (23.24) indicated that during the observation period, after reference to time decay, the ECG pattern and lung function maintained a high correlation.

[0427] formula:

[0428]

[0429] Analysis and examples:

[0430] Assume k is 2.0 (a constant that adjusts the response rate).

[0431] Calculation steps:

[0432] Application formula:

[0433]

[0434] Calculate e 11.62 , log(1+e 11.62 ), and then apply the above formula to get the result.

[0435] Result analysis:

[0436] The calculated fitness assessment result U is assumed to be approximately 95%, indicating a high cardiorespiratory fitness association, reflecting excellent health status.

[0437] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A health monitoring system based on the Internet of Things, characterized in that: The system comprises: The morphological feature capture module separates the QRS complex wave from the ECG signal and measures the morphological features including shape, width and height, applies opening and closing operations and morphological gradients to the morphological features, extracts waveform data, and obtains key morphological features; The ECG pattern analysis module is based on the key morphological features, serializes the ECG data and inputs it into the long short-term memory network, calculates the sample entropy and approximate entropy of the ECG signal cycle multiple times, aggregates the Lyapunov exponent of each ECG cycle, and analyzes it through phase space reconstruction technology to obtain the ECG behavior pattern; The dynamic analysis module of lung function inputs the patient's age, medical history and lung function data into the conditional variational autoencoder for feature extraction, inputs the extracted features into the bidirectional long short-term memory network, analyzes the dynamic changes of time dependence, and reconstructs and compares the current lung function data with the historical data to obtain the lung function variation characteristics; The data joint application module analyzes and compares the electrocardiographic behavior pattern and the lung function variation characteristics, analyzes the correlation and time series changes of the two data, evaluates the health status, and obtains the health assessment result.

2. The health monitoring system based on the Internet of Things according to claim 1, characterized in that: The steps for obtaining the key morphological features are specifically as follows: The Pan-Tompkins algorithm is used to locate the peak of the R wave in the ECG signal using the formula: ； Generate R wave peak position ; in, represents the peak position of the R wave, Represents the ECG signal, Represents time, represents the calibration factor; Using the R wave peak position , measure the width and height of the QRS complex wave, using the formula: and ; Generate QRS complex width and height QRS_dimensions; in, Represents the width, Represents height, , represent the time positions of the Q wave and the S wave, respectively. represents the time scaling factor, represents the amplitude magnification factor; The QRS complex width and height QRS_dimensions are applied to perform morphological operations, identify waveforms and extract features, using the formula: ； Obtain the key morphological feature G; in, Represents ECG signal The time derivative, , Represent the opening and closing operations respectively. Represents the normalized coefficient after operation.

3. The health monitoring system based on the Internet of Things according to claim 2, characterized in that: The steps for obtaining the sample entropy and approximate entropy of the ECG signal cycle are specifically as follows: Based on the key morphological features of ECG data, it is input into the long short-term memory network for processing, and the data serialization process is optimized using the formula: ； Generate processed serialized ECG data X; Among them, LSTM stands for long short-term memory network processing function, is the network tuning parameter, is the adjustment factor; The serialized ECG data is applied to calculate sample entropy and approximate entropy at the same time, and the accuracy of data analysis is enhanced through mathematical models, using the formula: and ; Get the sample entropy and approximate entropy of the ECG cycle; Among them, S represents sample entropy, A represents approximate entropy, represents the probability of the sequence, is the total number of subsequences, is the fitness calculation function, is the template length, is the goodness-of-fit threshold, is to adjust the weight, , is the normalization coefficient.

4. The health monitoring system based on the Internet of Things according to claim 3, characterized in that: The steps of acquiring the electrocardiographic behavior pattern are specifically as follows: Aggregate the sample entropy and approximate entropy of the ECG cycle to evaluate the behavioral characteristics of the overall ECG signal using the formula: ； Calculate the average value of the sum of sample entropy of the ECG signal period and the square root of the sum of squared approximate entropy to generate an entropy feature evaluation E of the ECG signal; in, It is The sample entropy of the period, It is The approximate entropy of the period, is the total entropy of the sample, is the total amount of approximate entropy; The entropy characteristic evaluation of the ECG signal is input into the phase space reconstruction technology to analyze the ECG behavior pattern using the formula: ； Generate electrocardiographic behavior pattern M; Among them, cos is the cosine function, represents the time delay coefficient, represents the entropy characteristics of the corresponding time point, is the total number of time points.

5. The health monitoring system based on the Internet of Things according to claim 4, characterized in that: The steps of feature extraction are specifically as follows: The patient's age, medical history, and lung function data are input into the encoder part of the conditional variational autoencoder, and the data is standardized using the formula: ； Generate standardized input data ; in, represents the original input data, represents the average value of the data, represents the standard deviation of the data, represents a constant to avoid division by zero errors; A weighted and biased nonlinear transformation is applied to the normalized input data using the formula: ； Calculate the latent variables and generate the latent space mapping results ; in, represents the weight of the encoder, represents the encoder bias, , is the adjustment parameter; The latent space mapping result is used to decode in the decoder, using the formula: ； Introducing fractions and absolute values ​​increases the nonlinear characteristics of the model, compares the reconstructed data with the original data, calculates the reconstruction error, and generates the reconstructed data ; in, represents the weight of the decoder, represents the bias of the decoder; The reconstructed data is normalized using the formula: ； Perform feature extraction ; in, represents the minimum value of the reconstructed data, represents the maximum value of the reconstructed data, Represents a constant that avoids a zero denominator.

6. The health monitoring system based on the Internet of Things according to claim 5, characterized in that: The steps for obtaining the lung function variation characteristics are specifically as follows: The feature extraction results are integrated with the historical lung function data set, and a composite calculation formula is used: ； Generate integrated dataset ; in, represents a historical lung function dataset, , represents the weight coefficient, represents the balance nonlinear adjustment coefficient; The bidirectional long short-term memory network is applied to process the integrated data set D and analyze the dynamic changes of time dependence using the formula: ； Generate time-dependent feature analysis results ; Among them, BiLSTM stands for bidirectional long short-term memory network, Adjust parameters for the network; The time-dependent features were compared with the historical lung function data set and the variation features were calculated using the formula: ； Generate a signature of lung function variation ; in, represents a historical lung function dataset, is the coefficient for adjusting the sensitivity of differences.

7. The health monitoring system based on the Internet of Things according to claim 6, characterized in that: The steps for obtaining the health assessment results are specifically as follows: From the time series data of the ECG behavior pattern and the lung function variability characteristics, the formula is used: ； Calculate and generate the normalized correlation coefficient R; in, represents the value of the i-th ECG behavior pattern, represents the value of the ith lung function variation feature, , represents the standard deviation of the ECG behavior pattern and lung function variability characteristics, represents a small constant to avoid division by zero; The normalized correlation coefficient is subjected to time window analysis using the formula: ； Calculate the weighted cumulative association ; in, represents the normalized correlation coefficient of time, represents the attenuation factor, represents the total observation time; The health assessment is performed based on the weighted cumulative association, using the formula: ； Generate health assessment results ; in, represents the base of natural logarithms, Represents a constant that adjusts the response rate.

Images