A method and system for monitoring and early warning of multiple vital signs data

Abstract

This invention relates to the field of sleep monitoring technology and discloses a method and system for monitoring and early warning of multiple vital signs data. The method addresses the problem that traditional wearable devices cannot characterize the internal structure of sleep by proposing a method for constructing a multi-scale heart rate third-order correlation index. This invention relies on heart rate (PPG) data collected by a smart bracelet, obtains the ultradian sleep rhythm component, the LC-NA ultraslow oscillation component, and the 0.1Hz vascular slow oscillation component through frequency domain decomposition, and constructs an instantaneous third-order correlation signal based on the standardized results of these three components. It further calculates the overnight global correlation index and the pseudo-periodic local correlation trajectory to form a comprehensive structural correlation index. This method can reflect the degree of coordination of the neural-autonomic-vascular system during sleep, accurately characterizing the stability of sleep structure, correlation strength, and its individualized deviation characteristics. This invention significantly improves the accuracy and practicality of smart bracelets in sleep quality assessment and long-term trend analysis.

Description

Technical Field

[0001] This invention relates to the field of sleep monitoring technology, and more specifically, to a method and system for monitoring and early warning of multiple vital signs. Background Technology

[0002] With the widespread adoption of smart wearable devices, heart rate monitoring based on photoplethysmography (PPG) has become the mainstream method for sleep analysis. Existing sleep monitoring algorithms mainly focus on time-domain statistics (such as heart rate variability (HRV)) and state classification (such as distinguishing between deep sleep, light sleep, REM sleep, and wakefulness). These methods are typically based on machine learning models, mapping extracted features to discrete sleep stage results, ultimately outputting macroscopic statistical indicators such as total sleep duration and the proportion of each stage. However, modern neurophysiological and sleep medicine research shows that healthy sleep is not a simple state transition, but a dynamic process involving a high degree of coordination among the central nervous system, autonomic nervous system, and cardiovascular system. Specifically, a long sleep rhythm of approximately 90 minutes controls the periodic alternation of NREM-REM sleep; the locus coeruleus-norepinephrine system-dominated neural ultraslow oscillations on the order of tens of seconds regulate micro-arousals and state transitions; and slow vascular oscillations with a frequency around 0.1 Hz are closely related to the pump effect of the cerebrospinal fluid clearance system. The main drawback of existing technologies is that they neglect the complete correlation between the above-mentioned multi-scale physiological processes. Traditional monitoring methods often treat heart rate signals as a single-dimensional time series, focusing only on their time-domain fluctuation amplitude or frequency-domain power spectral density. This fails to effectively separate the physiological oscillations at the three different time scales from a single-channel signal, making it difficult to characterize whether these three components maintain normal synchronization or resonance throughout the night's sleep. This results in current monitoring methods being unable to identify deeper, latent pathological states such as sufficient sleep duration but fragmented sleep structure, or prolonged deep sleep but neurovascular decoupling, and thus unable to provide early warnings regarding sleep structure stability. Summary of the Invention

[0003] This invention provides a method and system for monitoring and early warning of multiple vital signs data, which solves the technical problems mentioned in the background art.

[0004] Firstly, a multi-vital sign data monitoring and early warning system includes:

[0005] The data acquisition module is used to respond to the user activating the sleep monitoring function of the wearable device to obtain the heart rate time series submitted by the target object within the preset monitoring period.

[0006] The frequency domain decomposition module is used to perform multi-scale frequency domain decomposition on the heart rate time series and separate the first, second and third physiological components that respectively characterize the ultra-long sleep rhythm, the ultra-slow neural oscillation and the slow vascular oscillation.

[0007] The signal construction module is used to perform zero-mean standardization on the first, second and third physiological components respectively, and calculate the algebraic product of the standardized first, second and third physiological components at the same time to construct an instantaneous third-order correlation signal reflecting the synchronicity of multiple systems.

[0008] The statistical analysis module is used to perform time-domain statistical analysis on the instantaneous third-order correlation signal and calculate the multi-scale correlation index that quantifies the structured resonance intensity between different physiological systems.

[0009] The early warning generation module is used to generate sleep monitoring early warning information based on the comparison results between the multi-scale correlation index and the individual's historical baseline.

[0010] Secondly, a method for monitoring and early warning of multiple vital signs data, applied to any of the aforementioned monitoring and early warning systems for multiple vital signs data, includes:

[0011] In response to the user activating the sleep monitoring function of the wearable device, the heart rate time series submitted by the target subject within the preset monitoring period is obtained;

[0012] The heart rate time series was decomposed into multi-scale frequency domain to separate the first, second and third physiological components that respectively characterize the ultra-long sleep rhythm, the ultra-slow neural oscillation and the slow vascular oscillation.

[0013] The signal construction module is used to perform zero-mean standardization on the first, second and third physiological components respectively, and calculate the algebraic product of the standardized first, second and third physiological components at the same time to construct an instantaneous third-order correlation signal reflecting the synchronicity of multiple systems.

[0014] Time-domain statistical analysis was performed on the instantaneous third-order correlation signal to calculate the multi-scale correlation index that quantifies the structured resonance intensity between different physiological systems;

[0015] Based on the comparison results between the multi-scale correlation index and the individual's historical baseline, sleep monitoring and early warning information is generated.

[0016] The beneficial effects of this invention include: by constructing a multi-scale heart rate third-order correlation index, it achieves a quantitative characterization of sleep structural integrity that is impossible to obtain with existing smart bracelets. It can reconstruct and evaluate the correlation strength and stability between nighttime ultradian sleep rhythms, LC-NA ultraslow oscillations, and 0.1Hz vascular slow oscillations based on heart rate data collected by the bracelet, thereby revealing deep-seated neural-autonomic-vascular multi-scale coordinated states that are difficult to capture with traditional sleep monitoring methods. This invention can output continuous, personalized, and structured sleep correlation features, significantly improving the accuracy of consumer-grade wearable devices in sleep microstructure monitoring and long-term trend analysis. Attached Figure Description

[0017] Figure 1 This is a flowchart of a multi-vital sign data monitoring and early warning method according to the present invention;

[0018] Figure 2 This is a block diagram of a multi-vital sign data monitoring and early warning system according to the present invention;

[0019] Figure 3 This is a graph showing the evolution of heart rate and instantaneous third-order correlation signals according to the present invention. Detailed Implementation

[0020] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0021] Example 1: As Figure 1 As shown, a multi-vital sign data monitoring and early warning system includes:

[0022] The data acquisition module is used to respond to the user activating the sleep monitoring function of the wearable device to obtain the heart rate time series submitted by the target object within the preset monitoring period.

[0023] The frequency domain decomposition module is used to perform multi-scale frequency domain decomposition on the heart rate time series and separate the first, second and third physiological components that respectively characterize the ultra-long sleep rhythm, the ultra-slow neural oscillation and the slow vascular oscillation.

[0024] The signal construction module is used to perform zero-mean standardization on the first, second and third physiological components respectively, and calculate the algebraic product of the standardized first, second and third physiological components at the same time to construct an instantaneous third-order correlation signal reflecting the synchronicity of multiple systems.

[0025] The statistical analysis module is used to perform time-domain statistical analysis on the instantaneous third-order correlation signal and calculate the multi-scale correlation index that quantifies the structured resonance intensity between different physiological systems.

[0026] The early warning generation module is used to generate sleep monitoring early warning information based on the comparison results between the multi-scale correlation index and the individual's historical baseline.

[0027] Preferably, obtaining the heart rate time series submitted by the target subject within a preset monitoring period includes:

[0028] In response to the activation of the sleep monitoring function, the raw heart rate sequence is obtained by analyzing the photoplethysmography (PPG) signal. This is based on the start time of the preset monitoring period. and end time Construct an equally spaced target time grid sequence :

[0029]

[0030] in, : indicates the first The timestamp of each target sampling point : Represents the total number of sampling points, and , : Indicates the target sampling interval, preferably set to 1 second (corresponding to the sampling frequency) Hz).

[0031] For each target time point Locate the two actual sampling points before and after the original heart rate sequence. and ,satisfy The heart rate at that moment was calculated using a linear interpolation formula. :

[0032]

[0033] in, : indicates the first time interval mapped onto an equally spaced time grid. Heart rate values, : Represents the timestamp in the original sequence The corresponding measured heart rate value, : Represents the timestamp in the original sequence The corresponding measured heart rate value, , : Represents adjacent timestamps in the original non-equal interval sequence.

[0034] Using a preset length Sliding window for sequence Median filtering is performed to remove sharp impulse noise. The calculation formula is as follows:

[0035]

[0036] by For example, the specific calculation formula is as follows:

[0037]

[0038] in, : Indicates the th after the first stage of median filtering. An intermediate heart rate value, : Indicates the length of the median filter window, used to cover transient artifact impacts. : This indicates the operation of retrieving the median of the values ​​in the set.

[0039] Using a preset length Sliding window for sequence Perform average filtering to obtain the final preprocessed heart rate time series. The calculation formula is as follows:

[0040]

[0041] in, (Right now ): indicates the first digit after preprocessing. The heart rate time series values ​​are used as input for subsequent frequency domain decomposition. : Indicates the length of the moving average filter window, preferably set to 5, to achieve a low-pass filtering effect with a cutoff frequency of approximately 0.1Hz. : Represents the summation index variable within the sliding window.

[0042] The monitoring period starts at the time set by the user to begin sleep monitoring, preferably at 10 PM, as this is the normal time for most people to fall asleep, which aligns with the nighttime sleep monitoring scenario.

[0043] The end time of the monitoring period is the sleep monitoring termination time set by the user, preferably 6 o'clock, because this is the normal wake-up time for most people, ensuring coverage of the complete nighttime sleep cycle.

[0044] The target sampling interval is the time interval for mapping the original heart rate sequence to an equally spaced sequence, preferably 1 second, because this interval can balance data accuracy and device storage pressure, and meet the resolution requirements of subsequent frequency domain decomposition.

[0045] The sampling frequency is the frequency at which the wearable device collects photoplethysmography (PPG) signals, preferably 1 Hz, so as to correspond to the 1-second target sampling interval and ensure time grid alignment.

[0046] The sliding median filter window length is the number of data points in the window when performing median filtering, preferably 5, because this length can effectively remove instantaneous sharp impulse noise without destroying the trend characteristics of the heart rate signal.

[0047] The moving average filter window length is the number of window data points when performing average filtering, preferably 5, because this length can smooth signal fluctuations, achieve a low-pass filtering effect of about 0.1 Hz, and preserve the target frequency components.

[0048] The fixed sampling period is a fixed time interval for the wearable device to collect photoplethysmography (PPG) signals, preferably 1 second, so as to be consistent with the sampling frequency of 1 Hz and ensure the regularity of the original data collection.

[0049] The raw heart rate sequence is processed end-to-end by the wearable device. From photoplethysmography (PPG) signal analysis, data extraction, interpolation, to filtering and smoothing, all steps are completed within the wearable device itself, eliminating the need for interaction with external devices (such as phones or computers). This avoids data loss or delays during transmission, ensuring data security and integrity. For example, a certain brand of smart bracelet can complete all the above preprocessing operations locally without needing a real-time connection to a mobile phone.

[0050] In the combined filtering strategy, sharp impulse noise is first removed by moving median filtering, and then the signal is smoothed by moving average filtering. Specifically, sharp impulse noise is mostly caused by loose device wear or momentary limb movements. Median filtering can effectively remove such isolated outliers. Moving average filtering further smooths the signal by addressing the small fluctuations remaining after median filtering. For example, an isolated data point of 120 beats per minute that suddenly appears in the original heart rate sequence will be replaced by the median of adjacent normal data after 5-point median filtering, and then the signal curve will be smoother after 5-point moving average filtering.

[0051] The linear interpolation process is as follows: First, a target time grid with 1-second intervals is generated based on the start and end times. Then, for each target time point, two adjacent actual sampling points are found in the original non-equal interval heart rate sequence. The heart rate value at the target time point is obtained by calculating the linear relationship between the two points. For example, if the target time point is 22:00:00, and the heart rate in the original sequence is 60 beats per minute at 21:59:59 and 62 beats per minute at 22:00:02, then the heart rate at the target time point is calculated to be 61 beats per minute through linear interpolation, ensuring the continuity of the sequence.

[0052] The preferred value for a fixed sampling period is 1 second, because this value is consistent with the target sampling interval and sampling frequency, which can ensure that the time grid of the original data matches the time grid of subsequent processing, and is in line with the hardware performance of consumer-grade wearable devices, without causing excessive power consumption.

[0053] The preset lengths of the moving median filter window and the moving average filter window are not explicitly stated in the text. It is suggested that the optimal value for both is 5, as this length strikes a balance between noise removal effectiveness and signal fidelity. A window that is too short (e.g., 3 points) cannot effectively remove noise, while a window that is too long (e.g., 7 points) will cause signal lag and loss of key features. A 5-point window is the optimal choice as verified by experiments.

[0054] Preferably, the heart rate time series is decomposed into multiple scales in the frequency domain to separate the first physiological component characterizing the long sleep rhythm, specifically including:

[0055] Preprocessed heart rate time series Performing a Discrete Fourier Transform converts the signal from the time domain to the frequency domain, yielding a frequency-domain heart rate sequence. The calculation formula is as follows:

[0056]

[0057] in, : Indicates that the frequency index is The complex value in the frequency domain at that point, : indicates the preprocessed time-domain heart rate sequence at the 1st... Values ​​at each point in time. : Indicates the total number of sampling points, : Represents the imaginary unit, : Represents frequency index, , : indicates the first The actual frequency value corresponding to each index, and ,in The sampling frequency.

[0058] Constructing the first frequency domain filter for the ultra-low frequency band The filter is designed as an ideal low-pass filter, with its passband cutoff frequency set to [value missing]. Hertz. To ensure the conjugate symmetry of the Discrete Fourier Transform, the filter transfer function is defined as follows:

[0059]

[0060] in, : Indicates the first frequency domain filter at the frequency index Gain coefficient at that location, This indicates that the upper limit of the cutoff frequency corresponding to the periodic switching between non-REM sleep and REM sleep is retained, with a value of 0.004Hz. : Indicates the sampling frequency (e.g., 1Hz).

[0061] The frequency domain heart rate sequence is multiplied point-by-point with the first frequency domain filter to extract the frequency domain features of the first component. The calculation formula is as follows:

[0062]

[0063] in, : This represents a complex frequency domain sequence after extracting the energy of the extended sleep rhythm, retaining the energy from 0Hz to 0.004Hz in the frequency domain, while setting the frequency components above 0.004Hz to zero.

[0064] Frequency domain features of the first component Performing an inverse discrete Fourier transform yields the first physiological component in the time domain. The calculation formula is as follows:

[0065]

[0066] in, : Represents the first physiological component sequence in the time domain, characterizing the macroscopic baseline changes in the sleep ultra-long rhythm. : indicates taking the real part of the calculation result.

[0067] The first frequency domain filter passband frequency range is the frequency interval used to filter specific frequency components in the frequency domain heart rate sequence. It is preferably 0 Hz to 0.004 Hz, because this frequency range matches the periodic switching frequency of non-rapid eye movement sleep and rapid eye movement sleep, and can accurately retain the corresponding physiological rhythm components and eliminate other irrelevant frequency interference.

[0068] A dedicated first frequency domain filter is designed for the ultra-low frequency band. This filter is an ideal low-pass filter, allowing only frequency components from 0 Hz to 0.004 Hz to pass through, while completely suppressing other frequency components. The design of the ideal low-pass filter must adhere to the conjugate symmetry of the Discrete Fourier Transform to ensure that the signal after frequency domain processing can be accurately restored to the time domain signal through inverse transform. For example, in a frequency domain heart rate sequence, the 0.003 Hz frequency component will be retained, while the 0.005 Hz frequency component will be filtered out.

[0069] The Discrete Fourier Transform (DFT) is used to convert the time-domain heart rate sequence to the frequency domain. This transform can decompose the complex heart rate fluctuations in the time domain into sinusoidal wave components of different frequencies, each corresponding to a specific physiological activity. For example, a heart rate signal with smooth fluctuations in the time domain will exhibit frequency peaks of different amplitudes in the frequency domain after the DFT, making it easier to select target frequency components.

[0070] After frequency domain filtering, the time domain signal is restored through inverse discrete Fourier transform. This inverse transform can recombine the filtered frequency domain features into continuous time-domain physiological components, fully preserving the temporal evolution characteristics of the target physiological rhythm. For example, the filtered 0 Hz to 0.004 Hz frequency domain components can be restored into a time domain signal representing the ultra-long sleep rhythm after inverse transform, clearly reflecting the periodic alternation of non-rapid eye movement (NREM) sleep and rapid eye movement (REM) sleep.

[0071] The ultra-low frequency band is physiologically associated with the periodic switching between non-rapid eye movement (NREM) sleep and rapid eye movement (REM) sleep. By precisely locking onto this frequency band through a filter, the ultra-long sleep rhythm component can be separated, avoiding interference from frequency components generated by other physiological activities.

[0072] Preferably, the heart rate time series is decomposed into a multi-scale frequency domain to separate a second physiological component characterizing the ultraslow oscillation of the nerve, specifically including:

[0073] Construct a second frequency domain filter for the low-frequency band The filter is designed as an ideal bandpass filter, with its passband frequency range set to [value missing]. Hertz Hertz. To ensure that frequency domain operations are effective under the symmetry rules of the Discrete Fourier Transform, the filter transfer function is defined as follows:

[0074]

[0075] in, : Indicates the second frequency domain filter at the frequency index Gain coefficient at that location, : Indicates frequency index The corresponding actual frequency value, : This indicates that the lower cutoff frequency corresponding to the norepinephrine regulation of the locus coeruleus is retained, with a value of 0.01Hz. : Indicates the upper cutoff frequency, with a value of 0.04Hz. : Indicates the sampling frequency (e.g., 1Hz).

[0076] Using the frequency domain heart rate sequence By performing point-by-point multiplication with the second frequency domain filter, the frequency domain features of the second component are extracted. The calculation formula is as follows:

[0077]

[0078] in, : Represents a frequency domain complex sequence after extracting the energy of the neural ultraslow oscillation. : Represents a full-band heart rate frequency domain sequence.

[0079] Frequency domain features of the second component Performing an inverse discrete Fourier transform yields the second physiological component in the time domain. The calculation formula is as follows:

[0080]

[0081] in, : Represents the second physiological component sequence in the time domain, characterizing neural micro-awakening or state oscillations with a period of approximately 25 to 100 seconds. : Indicates the total number of sampling points, : Represents the imaginary unit, : indicates the operation of taking the real part, which yields the final real physical signal.

[0082] The passband frequency range of the second frequency domain filter is a frequency range used to screen specific low-frequency components in the frequency domain heart rate sequence. It is preferably 0.01 Hz to 0.04 Hz, because this frequency range matches the frequency of micro-arousal and sleep state transition regulated by the locus coeruleus norepinephrine (LC-NA) system, and can accurately retain the corresponding physiological oscillation components and eliminate other irrelevant frequency interference.

[0083] A dedicated second frequency domain filter is designed for the low-frequency band. This filter is an ideal bandpass filter, allowing only frequency components from 0.01 Hz to 0.04 Hz to pass through, while completely suppressing other frequency components. The design of an ideal bandpass filter must adhere to the conjugate symmetry of the Discrete Fourier Transform to ensure that the signal after frequency domain processing can be accurately restored to the time domain signal through inverse transform. For example, in a frequency domain heart rate sequence, the 0.02 Hz frequency component will be retained, while the 0.05 Hz frequency component will be filtered out.

[0084] This low-frequency band is physiologically associated with the locus coeruleus norepinephrine system, which is the core system for regulating sleep micro-arousals and state transitions. Its dominant physiological oscillation cycle is 25 to 100 seconds, and the corresponding frequency falls exactly in the range of 0.01 Hz to 0.04 Hz. By precisely locking this frequency band through a filter, the ultra-slow oscillation component of the nerve can be separated.

[0085] The discrete Fourier transform is used to convert the time-domain heart rate signal to the frequency domain. This transform can decompose the complex heart rate fluctuations in the time domain into sinusoidal components of different frequencies. Each component corresponds to a specific physiological activity, which makes it easier to select target frequency components.

[0086] After frequency domain filtering, the time domain signal is restored by inverse discrete Fourier transform. This inverse transform can recombine the selected frequency domain features into continuous time-domain physiological components, fully preserving the time evolution characteristics of neural ultra-slow oscillations. For example, the selected frequency domain components from 0.01 Hz to 0.04 Hz can be restored into time domain signals characterizing neural micro-awakening and state transitions after inverse transform.

[0087] Preferably, the heart rate time series is decomposed into a multi-scale frequency domain to separate a third physiological component characterizing slow vascular oscillations, specifically including:

[0088] Construct a third frequency domain filter for the slow oscillation band. The filter is designed as an ideal bandpass filter, with its passband frequency range set to [value missing]. Hertz Hertz. To ensure that frequency domain operations comply with the conjugate symmetry requirement of the Discrete Fourier Transform, the mathematical definition of the filter transfer function is as follows:

[0089]

[0090] in, : Indicates the third frequency domain filter at the frequency index Gain coefficient at that location, : Indicates frequency index The corresponding actual frequency value, : Indicates the retention of the lower cutoff frequency corresponding to vasomotor activity, with a value of 0.05Hz. : Indicates the upper cutoff frequency corresponding to the pump effect of the cerebrospinal fluid clearance system, with a value of 0.14Hz. : Indicates the sampling frequency (e.g., 1Hz).

[0091] Using the frequency domain heart rate sequence Perform point-by-point multiplication with the third frequency domain filter to extract the frequency domain features of the third component. The calculation formula is as follows:

[0092]

[0093] in, : Represents a complex frequency domain sequence after extracting the energy of the Mayer wave (slow oscillation in blood vessels). : Represents a full-band heart rate frequency domain sequence.

[0094] Frequency domain features of the third component Performing an inverse discrete Fourier transform yields the third physiological component in the time domain. The calculation formula is as follows:

[0095]

[0096] in, : Represents the third physiological component sequence in the time domain, characterizing vasomotor oscillations with a frequency around 0.1 Hz. : Indicates the total number of sampling points, : Represents the imaginary unit, : indicates the operation of taking the real part to obtain the final real physical signal.

[0097] The passband frequency range of the third frequency domain filter is the frequency range used to screen specific slow oscillation components in the frequency domain heart rate sequence. It is preferably 0.05 Hz to 0.14 Hz because this frequency range matches the pumping frequency of vasomotor activity and the cerebrospinal fluid clearance system, which can accurately retain the corresponding physiological oscillation components and eliminate other irrelevant frequency interference.

[0098] A dedicated third frequency domain filter is designed for the slow oscillation band. This filter is an ideal bandpass filter, allowing only frequency components from 0.05 Hz to 0.14 Hz to pass through, while completely suppressing other frequency components. The design of the ideal bandpass filter must adhere to the conjugate symmetry of the Discrete Fourier Transform to ensure that the signal after frequency domain processing can be accurately restored to the time domain signal through inverse transform. For example, in a frequency domain heart rate sequence, the 0.1 Hz frequency component will be retained, while the 0.04 Hz or 0.15 Hz frequency components will be filtered out.

[0099] The slow oscillation frequency band is associated with physiological mechanisms, including Mayer waves generated by vasomotor activity and the pumping effect of the cerebrospinal fluid clearance system. Its core frequency is concentrated around 0.1 Hz, which falls exactly in the range of 0.05 Hz to 0.14 Hz. By precisely locking this frequency band through a filter, the slow oscillation component of the blood vessels can be separated.

[0100] The discrete Fourier transform is used to convert the time-domain heart rate signal to the frequency domain. This transform can decompose the complex heart rate fluctuations in the time domain into sinusoidal components of different frequencies. Each component corresponds to a specific physiological activity, which makes it easier to select target frequency components.

[0101] After frequency domain filtering, the time domain signal is restored by inverse discrete Fourier transform. This inverse transform can recombine the selected frequency domain features into continuous time-domain physiological components, fully preserving the time evolution characteristics of slow vascular oscillations. For example, the selected frequency domain components from 0.05 Hz to 0.14 Hz can be restored to time domain signals characterizing the activity of the vasomotor and cerebrospinal fluid clearance systems after inverse transform.

[0102] Preferably, zero-mean standardization is performed on the first, second, and third physiological components, specifically including:

[0103] For the first physiological component respectively Second physiological component and the third physiological component Calculate its arithmetic mean over the entire preset monitoring period. and standard deviation Using any physiological component sequence For example, the calculation formula is as follows:

[0104] Arithmetic mean calculation:

[0105]

[0106] Standard deviation calculation:

[0107]

[0108] in, : Represents the physiological component sequences to be processed (respectively) , or ), : Indicates the total number of sampling points monitored throughout the night. : Represents the global arithmetic mean of this physiological component. : Represents the global standard deviation of this physiological component.

[0109] Using the calculated statistics, each physiological component sequence is standardized point-by-point to generate a standardized sequence with zero mean and unit variance. To prevent calculation errors caused by a zero standard deviation, a very small positive number is introduced. The calculation formula is as follows:

[0110]

[0111] Applying the above formulas, the standardized first physiological components were obtained respectively. Standardized second physiological component and the standardized third physiological component :

[0112]

[0113] in, : Represents the standardized physiological component values. : Represents a very small constant used to prevent the denominator from being zero (e.g. ), : These correspond to the arithmetic mean of the first, second, and third physiological components, respectively. : These correspond to the standard deviations of the first, second, and third physiological components, respectively;

[0114] Output , and All of them satisfy the statistical properties of a mean of 0 and a variance of 1, thus achieving dimensionless alignment of physiological signals with different amplitudes.

[0115] The minimum constant is an auxiliary value used to prevent calculation errors caused by the standard deviation being zero during the zero-mean standardization process. It is preferably one part per hundred thousand because this value is small enough that it will not affect the accuracy of the standardization results, and at the same time, it can effectively avoid the calculation anomaly of the denominator being zero.

[0116] Independent zero-mean standardization was performed on the first, second, and third physiological components, meaning that each component calculated its own arithmetic mean and standard deviation separately, without sharing statistical data with other components. For example, the first physiological component had a mean of 50 and a standard deviation of 5, while the second physiological component had a mean of 30 and a standard deviation of 3. Both were standardized based on their own data to ensure that the processing of each component was not affected by the amplitude characteristics of other components.

[0117] The specific process of zero-mean standardization is as follows: First, calculate the arithmetic mean of all data points for each physiological component throughout the entire monitoring period. Then, calculate the average of the sum of squares of the differences between all data points for that component and the mean. Finally, take the square root of this average to obtain the standard deviation. Subtract the mean of that component from the value of each data point to obtain the centered sequence. Then, divide each value in the centered sequence by the standard deviation of that component to obtain the standardized sequence. For example, if a data point for a physiological component has a value of 55, the mean of that component is 50, and the standard deviation is 5, then the centered value will be 5, and the standardized value will be 1.

[0118] By introducing a minimum constant and adding it to the standard deviation, we can avoid the situation where the standard deviation is zero due to all data points of a component having the same value, thus preventing meaningless division operations. For example, if all data points of a component are 40, its standard deviation is zero. After adding a minimum constant of one part in hundreds of thousands, the denominator becomes one part in hundreds of thousands, ensuring that the standardized calculation can be performed normally.

[0119] After standardization, all three physiological components have the characteristics of zero mean and unit variance, realizing dimensionless alignment of physiological signals with different amplitudes, allowing the three components with large amplitude differences to be processed on the same order of magnitude.

[0120] The preferred value to prevent the standard deviation from being zero is one in ten thousand, because this value can effectively avoid calculation anomalies without interfering with the standardization results.

[0121] Preferably, the algebraic product of the standardized first, second, and third physiological components at the same time is calculated to construct an instantaneous third-order correlation signal reflecting the synchronicity of multiple systems, specifically including:

[0122] Obtain the first physiological component sequence after zero-mean standardization. Second physiological component sequence and the third physiological component sequence ,in Ensure that the three sequences have the same sampling rate and phase reference on the time axis, i.e., the same index. Corresponding to the same physical moment or the corresponding moment after physiological delay correction.

[0123] Throughout the entire monitoring period, the three standardized physiological components were calculated point by point at the same time. The numerical product yields the instantaneous third-order correlation signal. The calculation formula is as follows:

[0124]

[0125] The generated sequence Defined as a quantitative indicator characterizing whether structured resonance occurs in the time domain of physiological processes at three different time scales during sleep: neural, autonomic, and vascular.

[0126] when When the value is significantly positive, it indicates that the three systems have resonated in the same direction at that moment (i.e., they are simultaneously in a significantly activated state above the baseline, or simultaneously in a significantly inhibited state below the baseline).

[0127] when When the value is close to zero, it indicates that at least one system is at the baseline level or that the systems are in a random state of non-association.

[0128] The timeline alignment operation is specifically based on a unified sampling rate and timestamp benchmark, ensuring that each data point index of the three standardized physiological component sequences corresponds to the same physical moment. For example, when the sampling rate is 1 Hz, index 0 corresponds to the monitoring start time, index 1 corresponds to 1 second after the start, and so on, ensuring that the values ​​of the three components at the same moment can be accurately matched and calculated.

[0129] Calculating the algebraic product point by point involves iterating through each time point in the entire monitoring period and directly multiplying the values ​​of the three standardized components at that moment without any additional weighting or correction. For example, if the first standardized component is 0.8, the second standardized component is 0.9, and the third standardized component is 1.0 at a certain moment, then the instantaneous third-order correlation signal value at that moment is 0.8 multiplied by 0.9 multiplied by 1.0, which equals 0.72.

[0130] The physiological significance of instantaneous third-order correlation signals is that when the value is significantly positive, it indicates that the three systems are simultaneously above or below their respective baseline levels, indicating a state of coordinated resonance; when the value is close to zero, it may mean that one component is at the baseline level, or that the three systems are fluctuating in different directions, indicating a state of non-correlation. For example, when sleep depth is good, the signal will show multiple significant positive peaks; when sleep is fragmented, the signal will mostly fluctuate near the zero line.

[0131] Preferably, time-domain statistical analysis is performed on the instantaneous third-order correlation signal to calculate a multi-scale correlation index that quantifies the structured resonance intensity between different physiological systems, specifically including:

[0132] Calculate the instantaneous third-order correlation signal The arithmetic mean over the entire preset monitoring period is used as the basic correlation strength. The calculation formula is as follows:

[0133]

[0134] in, : Represents the average neurovascular resonance level throughout the night. : Indicates the total number of sampling points, : indicates the first The instantaneous third-order correlation signal value at each moment.

[0135] Utilizing the preset ultra-long rhythm cycle length (Set to 5400 seconds, or 90 minutes) The entire monitoring period is divided into... Calculate the number of complete time segments. :

[0136]

[0137] Regarding the first Time segments (in ), calculate within this segment Local average This results in a periodic correlation strength sequence that evolves over time. :

[0138]

[0139] in, : Indicates the sampling interval (e.g., 1 second). : indicates the first The set of time indices corresponding to each segment. : indicates the first The total number of data points contained in each segment : indicates by The sequence formed.

[0140] For the periodic correlation strength sequence Perform statistical analysis and calculate the arithmetic mean of the sequences respectively. , sequence standard deviation and sequence skewness :

[0141] Sequence arithmetic mean:

[0142]

[0143] Sequence standard deviation:

[0144]

[0145] Sequence skewness:

[0146]

[0147] in, : Characterizing the average correlation height of sleep cycles, Characterizes the stability (fluctuation) of association strength during the sleep cycle. : Characterizes the distribution bias of association strength on the time axis (positive skewness indicates that strong associations are concentrated in the first half of the night). : A very small constant used to prevent division by zero errors.

[0148] Using preset weighting coefficients The above statistical characteristics are linearly weighted and summed to obtain the final multi-scale correlation index representing the integrity of sleep structure. The calculation formula is as follows:

[0149]

[0150] in, : Represents the final output quantization index, Average intensity weight (preferred to be 1.0) is the most direct indicator of sleep quality. Stability weight (preferably 0.5) is used as a deduction item; the greater the fluctuation, the lower the score. Distribution weight (preferred to be 0.2) rewards a distribution pattern that conforms to physiological laws (deep first, then shallow).

[0151] The length of the ultra-long rhythm cycle is a time benchmark used to divide the monitoring period. It is preferably 5400 seconds or 90 minutes because this duration matches the regular cycle of the ultra-long sleep rhythm and can accurately correspond to the periodic alternation of non-rapid eye movement sleep and rapid eye movement sleep.

[0152] The average intensity weight is a weighting coefficient for the arithmetic mean of the periodic correlation intensity sequence, preferably 1.0, because this mean directly reflects the core level of sleep multisystem association and is a key indicator for assessing sleep structure integrity.

[0153] The stability weight is a weighting coefficient for the standard deviation of the periodic correlation strength sequence, preferably 0.5, because the standard deviation reflects the degree of fluctuation in the correlation strength and can be used as a deduction item to balance the stability evaluation of the sleep structure.

[0154] The distribution weight is a weighting coefficient for the skewness of the periodic association strength sequence, preferably 0.2, because the skewness reflects the temporal distribution characteristics of the association strength, and as a reward, it can highlight the sleep structure that conforms to physiological laws.

[0155] The monitoring sessions are divided into 90-minute segments because this duration corresponds to a very long sleep rhythm cycle, and each segment can completely cover one alternation process between non-rapid eye movement (NREM) sleep and rapid eye movement (REM) sleep. For example, a monitoring duration of 720 minutes can be divided into 8 complete segments, each segment lasting 5400 seconds, to ensure that the segments match the physiological rhythm.

[0156] The construction logic of the periodic association strength sequence is to calculate the local average value of the instantaneous third-order association signal within each 90-minute segment, forming a sequence that evolves over time. For example, if the average instantaneous signal value is 0.3 in one segment and 0.2 in another, the sequence can intuitively reflect the changes in association strength across different sleep cycles.

[0157] We extracted three features from the sequences: arithmetic mean, standard deviation, and skewness, which characterize the association strength from three dimensions: core level, stability, and distribution pattern. The mean reflects the overall association degree, the standard deviation reflects the magnitude of fluctuation, and the skewness reflects the distribution bias. Positive skewness indicates that the strong association is concentrated in the first half of the night, which is consistent with the normal physiological pattern of sleep.

[0158] In the weighted summation formula, the mean is weighted and included positively, the standard deviation is weighted and included negatively, and the skewness is weighted and included positively. For example, if a sequence has a mean of 0.4, a standard deviation of 0.1, and a skewness of 0.3, after weighting, the multi-scale correlation index can comprehensively reflect the integrity of sleep structure and avoid the one-sidedness of evaluation by a single indicator.

[0159] The preferred length of the ultra-long circadian rhythm cycle is 5400 seconds, or 90 minutes. Since the normal cycle of ultra-long sleep circadian rhythms can be directly divided into time periods by those skilled in the art, this value can be used to ensure that the segmentation is consistent with the physiological rhythm.

[0160] The average intensity weight is preferably 0.5, the stability weight is preferably 0.3, and the distribution weight is preferably 0.2. Because of the importance of each feature to the evaluation of sleep structure, this value has been experimentally verified to balance the evaluation of each dimension and ensure the rationality and operability of the index.

[0161] Preferably, sleep monitoring early warning information is generated based on the comparison results between the multi-scale correlation index and the individual's historical baseline, specifically including:

[0162] Retrieve the target object's data before the current monitoring period. The multi-scale correlation index records calculated within a historical effective monitoring period constitute the historical sample set. Calculate the historical arithmetic mean of the historical sample set. and historical sample standard deviation The calculation formula is as follows:

[0163] Historical arithmetic mean:

[0164]

[0165] Historical sample standard deviation:

[0166]

[0167] in, : Indicates the length of the historical backtracking window (preferably 14, i.e., the past two weeks). : Indicates the index of the historical record. , : indicates the first The multi-scale correlation index is calculated from historical monitoring periods.

[0168] The multi-scale correlation index calculated during the current monitoring Subtracting the historical arithmetic mean and then dividing by the historical sample standard deviation (introducing a very small constant to prevent division by zero) yields a standard score reflecting the degree of deviation of the current sleep structure from the individualized baseline. The calculation formula is as follows:

[0169]

[0170] in, : Represents the dimensionless standardized deviation score (Z-score). : Represents the multi-scale correlation index calculated during the current monitoring period. : Represents a minimal constant to prevent the denominator from being zero (e.g. ).

[0171] The standard score With the preset score threshold Perform numerical comparisons and generate corresponding feedback signals based on the comparison results:

[0172] Decision logic:

[0173]

[0174] in, : Represents the statistical threshold for determining abnormality, preferably set to -1.0 (corresponding to approximately 16% quantiles on the lower side of a normal distribution). This indicates that the level of neurovascular synergy that night was significantly lower than the individual's normal level, triggering an early warning.

[0175] The historical backtracking window length is the number of valid historical monitoring data of the target object selected, preferably 14, because two weeks of monitoring data can fully reflect the normal level of individual sleep structure and provide reliable samples for building a baseline model.

[0176] The abnormality judgment score threshold is the critical value for judging whether the current sleep structure is abnormal. It is preferably -1.0 because this value corresponds to the 16th percentile on the lower side of the normal distribution, which can reasonably distinguish between normal fluctuations and significant abnormalities.

[0177] The logic behind constructing an individualized baseline model is based on the target object's own historical data, rather than a general group standard. A specific baseline is formed by calculating the historical arithmetic mean and standard deviation of the multi-scale correlation index over the past 14 monitoring periods, thus avoiding misjudgments caused by individual differences. For example, if person A has a historical mean of 0.6 and a standard deviation of 0.1, while person B has a historical mean of 0.4 and a standard deviation of 0.08, their baselines are different, and therefore their judgment criteria will also differ.

[0178] The standard score (Z-score) is calculated by subtracting the historical arithmetic mean from the current multiscale correlation index, and then dividing by the historical standard deviation, introducing a very small constant to avoid a zero denominator. This score can convert the current data into a dimensionless deviation from the baseline. For example, if the current index is 0.4, the historical mean is 0.6, the standard deviation is 0.1, and the standard score is -2.0, intuitively reflecting the degree of deviation.

[0179] The statistical basis for the warning rule is as follows: when the standard score is less than -1.0, it indicates that the current neurovascular synergy is significantly lower than the individual's normal level, triggering a warning; otherwise, it is determined that the sleep is sufficient. For example, if the standard score is -1.2, which is below the threshold, a sleep deficiency signal is output; if the standard score is 0.3, which is above the threshold, a sleep sufficient signal is output.

[0180] The anomaly detection score threshold is preferably -1.0, because this value conforms to the statistical characteristics of a normal distribution, which can effectively distinguish between normal and abnormal situations, ensuring the rationality and consistency of the warning.

[0181] like Figure 3 As shown, Figure 3 The evolution of heart rate data and instantaneous third-order correlation signals was compared to demonstrate the synchronicity characteristics of multiple physiological systems. The gray dashed line in the figure represents preprocessed heart rate data, exhibiting a smooth, periodic fluctuation of approximately 50 to 60 beats per minute, reflecting the macroscopic heart pumping rate. The black solid line represents the instantaneous third-order correlation signal, constructed by decomposing the raw heart rate into three independent components representing sleep rhythm, neural ultra-slow oscillations, and vascular slow oscillations, and calculating their instantaneous product. The significant spike in the black solid line indicates that the three physiological systems deviate in the same direction at the same moment, resulting in multi-scale resonance, reflecting the high degree of coordination between sleep rhythm, neural regulation, and vascular activity. The flatness of the signal near the zero line represents mutual cancellation or functional decoupling between the systems. This captures deep-seated structured resonance events that cannot be reflected in conventional heart rate monitoring, and based on the characteristic that system decoupling precedes overt heart rate abnormalities, it provides an early warning basis for declining sleep quality and health risks.

[0182] Example 2: A method for monitoring and early warning of multiple vital signs data, applied to any of the multiple vital signs data monitoring and early warning systems described in this embodiment, comprising:

[0183] In response to the user activating the sleep monitoring function of the wearable device, the heart rate time series submitted by the target subject within the preset monitoring period is obtained;

[0184] The heart rate time series was decomposed into multi-scale frequency domain to separate the first, second and third physiological components that respectively characterize the ultra-long sleep rhythm, the ultra-slow neural oscillation and the slow vascular oscillation.

[0185] The signal construction module is used to perform zero-mean standardization on the first, second and third physiological components respectively, and calculate the algebraic product of the standardized first, second and third physiological components at the same time to construct an instantaneous third-order correlation signal reflecting the synchronicity of multiple systems.

[0186] Time-domain statistical analysis was performed on the instantaneous third-order correlation signal to calculate the multi-scale correlation index that quantifies the structured resonance intensity between different physiological systems;

[0187] Based on the comparison results between the multi-scale correlation index and the individual's historical baseline, sleep monitoring and early warning information is generated.

[0188] Example 3: Implementation of monitoring and early warning based on heart rate data from multiple day and night scenarios combined with sleep rhythm.

[0189] For a complete 24-hour cycle, covering seven typical time periods: morning rest, after breakfast, morning work, after lunch, afternoon moderate-intensity exercise, after dinner relaxation, and nighttime sleep, photoplethysmography (PPG) heart rate data of the target subject (30-year-old adult male with regular daily exercise and no underlying diseases) is collected. By integrating multi-scene data from day and night with nighttime sleep rhythm through the original patented method, more comprehensive physiological coordination state monitoring and early warning can be achieved.

[0190] Data acquisition parameters;

[0191] Data collection device: Smart bracelet (supports 24-hour PPG sampling), fixed sampling period of 1 second (sampling frequency of 1 Hz).

[0192] Monitoring period: 24 hours (00:00-24:00), of which the nighttime sleep period is 23:00-07:00 (preset monitoring start time 23:00, end time 07:00).

[0193] Scene labeling: Record the start and end times of each period through the manual labeling function of the wristband (such as 07:30-08:00 after breakfast, 16:00-17:00 during exercise).

[0194] Multi-scenario data preprocessing;

[0195] Optimize preprocessing strategies for heart rate characteristics in different daytime scenarios:

[0196] Resting periods (morning, work, relaxation): Use 5-point moving median filter + 5-point moving average filter to remove minor body movement noise.

[0197] During exercise (16:00-17:00, jogging): the sliding median filter window length is adjusted to 7 to enhance the removal of sharp impulse noise generated by strenuous exercise; the sliding average filter window remains at 5 to preserve the core trend of heart rate changes during exercise.

[0198] Post-meal periods (after breakfast, lunch, and dinner): missing data caused by loose wristbands during the meal process are supplemented by linear interpolation to ensure the continuity of the time grid (target sampling interval of 1 second).

[0199] Data segmentation: Data from each of the seven marked time periods is extracted from the 24-hour continuous heart rate sequence, preprocessed separately, and then integrated into a complete 24-hour preprocessed heart rate time series.

[0200] Decompose parameters;

[0201] First physiological component (long sleep rhythm): 0-0.004 Hz (representing macroscopic energy changes in the diurnal rhythm during the daytime period);

[0202] The second physiological component (neural ultra-slow oscillation): 0.01-0.04 Hz (characterized by daytime mood and stress-related microstate transitions regulated by the locus coeruleus-norepinephrine system);

[0203] The third physiological component (slow vascular oscillation): 0.05-0.14 Hz (characterizing the dynamic response of vasomotor function to diet and exercise during the day).

[0204] Component extraction results;

[0205] Sleep period (23:00-07:00): The first physiological component shows a clear 90-minute cycle fluctuation, corresponding to the alternation of non-rapid eye movement sleep and rapid eye movement sleep; the second component fluctuates gently during deep sleep and the frequency of fluctuation increases during light sleep; the third component is synchronized with the activity of the cerebrospinal fluid clearance system.

[0206] During exercise (16:00-17:00): The fluctuation amplitude of the second component increases significantly, reflecting the real-time response of neural regulation to exercise intensity; the frequency of the third component increases slightly, corresponding to vasodilation to meet the blood oxygen demand during exercise.

[0207] In the postprandial period (e.g., 12:30-13:30 after lunch): the third component fluctuates more, reflecting the changes in vasomotor activity caused by digestive system activity; the second component fluctuates briefly before returning to stability, reflecting the rapid adaptation of the nervous system to the postprandial state.

[0208] During rest periods (such as 9:00-12:00 in the morning): the fluctuations of the three components are relatively smooth, while the second component occasionally fluctuates slightly, corresponding to the shift in attention during work.

[0209] Signal construction;

[0210] Zero-mean standardization was performed on the three physiological components extracted from each time period (introducing a very small constant of one in several hundred thousand to prevent calculation errors). After aligning the time axis, the algebraic product was calculated point by point to obtain a continuous 24-hour instantaneous third-order correlation signal.

[0211] Signal feature analysis;

[0212] During sleep: Significantly positive signal peaks are concentrated in the deep sleep stage, indicating a high degree of coordination between the neuro-autonomic-vascular system; during the light sleep stage, the signal fluctuates around the zero line, and the coordination is weakened.

[0213] During exercise: The signal shows a brief but significant positive peak, corresponding to the three systems responding synchronously to the body's needs during exercise; after exercise, the signal quickly drops back to a stable level.

[0214] Postprandial period: The signal shows a small positive fluctuation, which lasts for about 30 minutes before returning to a stable state, reflecting the coordinated regulation of digestion and metabolism by multiple systems after eating.

[0215] Resting period: The signal is generally close to zero, with occasional small fluctuations, indicating that the system is in a low-coordination basic maintenance state.

[0216] Calculation of multi-scale correlation index;

[0217] Use the following parameters:

[0218] The length of the ultra-long rhythm cycle is 5400 seconds (90 minutes, with daytime periods used to divide different activity cycles).

[0219] Weighting coefficients: average intensity weight 1.0, stability weight 0.5, distribution weight 0.2.

[0220] Calculate separately:

[0221] 24-hour global multi-scale correlation index (representing the overall level of day-night coordination);

[0222] Local multi-scale correlation indices for each time period (representing the collaborative state under a single scenario);

[0223] Specific correlation index for nighttime sleep periods (consistent with the calculation results of the original patent).

[0224] Baseline construction;

[0225] Retrieve 24-hour monitoring data of the target object over the past 14 days and construct the following:

[0226] Baseline of global correlation index for day and night (historical arithmetic mean + historical standard deviation).

[0227] Baseline of sleep duration-specific correlation index;

[0228] Baseline of local correlation index in key scenarios (exercise, post-eating, rest).

[0229] Comparison and early warning results;

[0230] Global comparison: The global correlation index for the current 24-hour period is 0.32, the historical average is 0.30, the standard deviation is 0.05, and the standard score is 0.4, which is higher than the threshold of -1.0. There is no global collaborative anomaly warning.

[0231] Scene comparison: The local correlation index during exercise was 0.28, with a historical average of 0.25, a standard deviation of 0.03, and a standard score of 1.0, indicating good system coordination during exercise; the local correlation index after dinner was 0.15, with a historical average of 0.22, a standard deviation of 0.04, and a standard score of -1.75, which was below the threshold, triggering an abnormal system coordination warning after dinner.

[0232] Sleep comparison: The sleep period-specific correlation index was 0.41, the historical average was 0.38, the standard deviation was 0.06, the standard score was 0.5, and the output signal showed an intact sleep structure.

[0233] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A monitoring and early warning system for multiple vital signs data, characterized in that, include: The data acquisition module is used to respond to the user activating the sleep monitoring function of the wearable device to obtain the heart rate time series submitted by the target object within the preset monitoring period. The frequency domain decomposition module is used to perform multi-scale frequency domain decomposition on the heart rate time series. After inverse discrete Fourier transform, it separates the first, second, and third physiological components that represent the ultra-long sleep rhythm, the ultra-slow neural oscillation, and the slow vascular oscillation, respectively. The frequency range of the first physiological component is between 0 Hz and 0.004 Hz, the frequency range of the second physiological component is between 0.01 Hz and 0.04 Hz, and the frequency range of the third physiological component is between 0.05 Hz and 0.14 Hz. The signal construction module is used to perform zero-mean standardization on the first, second and third physiological components respectively, and calculate the algebraic product of the standardized first, second and third physiological components at the same time to construct an instantaneous third-order correlation signal reflecting the synchronicity of multiple systems. The statistical analysis module is used to perform time-domain statistical analysis on the instantaneous third-order correlation signal and calculate the multi-scale correlation index that quantifies the structured resonance intensity between different physiological systems. The early warning generation module is used to generate sleep monitoring early warning information based on the comparison results between the multi-scale correlation index and the individual's historical baseline.

2. The monitoring and early warning system for multiple vital signs data according to claim 1, characterized in that, Obtain the heart rate time series submitted by the target object within a preset monitoring period, including: The wearable device collects photoplethysmography (PPG) signals at a fixed sampling period and analyzes them to obtain the raw heart rate sequence. Based on the system clock and the sleep monitoring time period set by the target object, nighttime data is extracted from the raw heart rate sequence. The extracted data undergoes linear interpolation and is mapped onto an equally spaced time grid to form a continuous time series. A preset-length sliding median filter window is used to filter the continuous time series to remove sharp impulse noise, followed by smoothing using a preset-length sliding average filter window to obtain a preprocessed heart rate time series. The raw heart rate sequence is processed end-to-end by the wearable device.

3. The monitoring and early warning system for multiple vital signs data according to claim 2, characterized in that, Multi-scale frequency domain decomposition was performed on the heart rate time series to separate the first physiological component characterizing the ultralong sleep rhythm, including: A discrete Fourier transform is performed on the preprocessed heart rate time series to convert it from the time domain to the frequency domain, resulting in a frequency domain heart rate series. A first frequency domain filter is constructed for the ultra-low frequency band, with its passband frequency range set between 0 Hz and 0.004 Hz to retain frequency components corresponding to the periodic switching between non-rapid eye movement (NREM) sleep and rapid eye movement (REM) sleep. The frequency domain heart rate series is multiplied by the first frequency domain filter to extract the frequency domain features of the first component. An inverse discrete Fourier transform is performed on the frequency domain features of the first component to restore the first physiological component in the time domain.

4. The monitoring and early warning system for multiple vital signs data according to claim 3, characterized in that, Multi-scale frequency domain decomposition was performed on the heart rate time series to separate the second physiological component characterizing the ultraslow oscillations of the nerve, including: A second frequency domain filter is constructed for the low-frequency band, with a passband frequency range set between 0.01 Hz and 0.04 Hz to retain the frequency components corresponding to micro-arousal and state transitions regulated by norepinephrine in the locus coeruleus. The frequency domain heart rate sequence is multiplied with the second frequency domain filter to extract the frequency domain features of the second component. The frequency domain features of the second component are subjected to an inverse discrete Fourier transform to restore the second physiological component in the time domain.

5. A monitoring and early warning system for multiple vital signs data according to claim 4, characterized in that, Multi-scale frequency domain decomposition was performed on the heart rate time series to separate the third physiological component characterizing slow vascular oscillations, including: A third frequency domain filter is constructed for the slow oscillation frequency band. The passband frequency range of this third frequency domain filter is set between 0.05 Hz and 0.14 Hz to retain the frequency components corresponding to vasomotor activity and the pump effect of the cerebrospinal fluid clearance system. The frequency domain features of the third component are extracted by multiplying the frequency domain heart rate sequence with the third frequency domain filter. The frequency domain features of the third component are then subjected to an inverse discrete Fourier transform to restore the third physiological component in the time domain.

6. The monitoring and early warning system for multiple vital signs data according to claim 5, characterized in that, Zero-mean standardization was performed on the first, second, and third physiological components, including: The arithmetic mean and standard deviation of the first physiological component, the second physiological component, and the third physiological component are calculated over the entire preset monitoring period. The value of each physiological component at each time point is subtracted from the arithmetic mean of that physiological component to obtain a centered sequence. Each value in the centered sequence is divided by the standard deviation of that physiological component to obtain three standardized physiological component sequences with zero mean and unit variance characteristics, respectively.

7. A monitoring and early warning system for multiple vital signs data according to claim 6, characterized in that, The algebraic product of the standardized first, second, and third physiological components at the same time point is calculated to construct an instantaneous third-order correlation signal reflecting the synchronicity of multiple systems, including: The standardized first, second, and third physiological component sequences are aligned on the time axis; the numerical products of the first, second, and third physiological components at the same time are calculated point by point; the resulting product sequence is used as an instantaneous third-order correlation signal characterizing whether physiological processes at three different time scales have undergone co-directional resonance shift in the time domain.

8. A monitoring and early warning system for multiple vital signs data according to claim 7, characterized in that, Time-domain statistical analysis was performed on the instantaneous third-order correlation signal to calculate a multi-scale correlation index that quantifies the structured resonance intensity between different physiological systems, including: The arithmetic mean of the instantaneous third-order correlation signal over the entire preset monitoring period is calculated as the basic correlation strength. The entire preset monitoring period is divided into several continuous time segments using a preset ultra-long rhythm cycle length. The local average value of the instantaneous third-order correlation signal in each time segment is calculated to form a periodic correlation strength sequence that evolves over time. The sequence arithmetic mean, sequence standard deviation, and sequence skewness of the periodic correlation strength sequence are calculated. The sequence arithmetic mean, sequence standard deviation, and sequence skewness are weighted and summed using preset weighting coefficients to obtain a multi-scale correlation index characterizing the integrity of sleep structure.

9. A monitoring and early warning system for multiple vital signs data according to claim 8, characterized in that, Based on the comparison results between the multi-scale correlation index and the individual's historical baseline, sleep monitoring early warning information is generated, including: The system retrieves historical data of the multi-scale correlation index calculated for the target object over several monitoring periods; calculates the historical arithmetic mean and standard deviation of the historical data to establish an individualized baseline model for the target object; subtracts the historical arithmetic mean from the multi-scale correlation index calculated in the current monitoring period, and then divides it by the historical standard deviation to obtain a standard score reflecting the degree of deviation of the current sleep structure from the individualized baseline; compares the standard score with a preset score threshold; if the standard score is greater than or equal to the preset score threshold, a sleep sufficiency signal is output through the wearable device; otherwise, a sleep deprivation signal is output through the wearable device.

10. A method for monitoring and early warning of multiple vital signs data, applied to a monitoring and early warning system for multiple vital signs data as described in any one of claims 1-9, characterized in that, include: In response to the user activating the sleep monitoring function of the wearable device, the heart rate time series submitted by the target subject within the preset monitoring period is obtained; The heart rate time series was decomposed into multiple scales in the frequency domain and subjected to inverse discrete Fourier transform to separate the first, second, and third physiological components that characterize the ultralong sleep rhythm, the ultraslow neural oscillation, and the slow vascular oscillation, respectively. The frequency range of the first physiological component is between 0 Hz and 0.004 Hz, the frequency range of the second physiological component is between 0.01 Hz and 0.04 Hz, and the frequency range of the third physiological component is between 0.05 Hz and 0.14 Hz. Zero-mean standardization was performed on the first, second, and third physiological components respectively, and the algebraic product of the standardized first, second, and third physiological components at the same time was calculated to construct an instantaneous third-order correlation signal reflecting the synchronicity of multiple systems. Time-domain statistical analysis was performed on the instantaneous third-order correlation signal to calculate the multi-scale correlation index that quantifies the structured resonance intensity between different physiological systems; Based on the comparison results between the multi-scale correlation index and the individual's historical baseline, sleep monitoring and early warning information is generated.

Images