A personalized sleep health risk assessment method based on 5G edge computing

Abstract

The present application relates to the technical field of health risk assessment, in particular to a personalized sleep health risk assessment method based on 5G edge computing, comprising the following steps: collecting multi-source sleep data through 5G edge computing and synchronously combining, extracting key features to generate a parameter set, calculating the period drift difference to analyze the parameter change trend, dynamically adjusting the monitoring interval according to the drift direction, constructing a dynamic health baseline model to calculate the risk index, fusing the multi-parameter risk coefficient to divide the risk level, fitting the continuous period trend to generate the optimized personalized sleep health risk assessment result, in the present application, through time synchronization and normalization of multi-source physiological and environmental data, data consistency and high-precision analysis are realized, through period drift difference, health change trend is identified, the monitoring interval is dynamically adjusted to improve flexibility and real-time performance, through steady-state parameter distribution and risk weighting calculation, the accuracy of sleep risk identification and individualized early warning ability are enhanced.

Description

Technical Field

[0001] This invention relates to the field of health risk assessment technology, and in particular to a personalized sleep health risk assessment method based on 5G edge computing. Background Technology

[0002] The field of health risk assessment technology encompasses technologies that utilize information and communication technologies and data analysis methods to monitor, analyze, and predict human health status. Its core lies in establishing health status models by collecting individual physiological parameters, behavioral characteristics, and environmental data to identify potential health hazards and risks. Its technical system typically covers health data acquisition, health indicator calculation, risk modeling, and predictive analysis, and is widely applied in personal health management, medical decision support, and remote health monitoring. Its development relies on the collaborative application of multi-source data fusion, intelligent analysis models, and communication network technologies, enabling health risk identification to extend from traditional medical testing to real-time, dynamic monitoring.

[0003] Among them, the personalized sleep health risk assessment method based on 5G edge computing refers to a health risk assessment technology that utilizes 5G communication networks and edge computing architecture to collect and analyze sleep data of family members. Targeting sleep behavior monitoring and risk assessment in a home setting, it primarily involves deploying data computing units at edge nodes to extract local features and calculate indicators from raw data such as sleep posture, respiratory rate, heart rate changes, and environmental noise. Risk assessment analysis is then conducted based on individual health parameters and existing risk assessment models. 5G communication enables low-latency data transmission and multi-source data collaborative processing to complete the calculation and assessment process for sleep-related health risks.

[0004] In existing technologies, health risk assessment typically relies on fixed time intervals or static monitoring ranges for data collection and analysis. This method is susceptible to inconsistencies in the frequency of physiological parameter changes or time deviations, resulting in low timeliness and accuracy of the monitoring data. Furthermore, existing technologies have limitations in real-time analysis of health data, often failing to capture short-term fluctuations and sudden changes in an individual's health status. For example, fixed monitoring ranges may miss health abnormalities that occur within a short period, causing the risk assessment to fail to reflect the individual's current health status in a timely manner. More importantly, existing technologies typically cannot dynamically adjust the monitoring range according to actual conditions, potentially leading to the omission of crucial data and affecting the final health risk assessment. These shortcomings prevent traditional health monitoring methods from meeting the needs for personalization and real-time monitoring, and from effectively improving the accuracy and flexibility of individual health management. Summary of the Invention

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a personalized sleep health risk assessment method based on 5G edge computing, comprising the following steps:

[0006] S1: Collect the average heart rate, respiratory rate, number of nighttime body movements, skin temperature changes, bedroom temperature and humidity, noise intensity and air quality of an individual during the sleep cycle through a 5G edge computing terminal, and perform time synchronization and normalization, extract key feature components, and generate a multi-dimensional set of key sleep feature parameters.

[0007] S2: Based on the multi-dimensional sleep key feature parameter set, calculate the cycle drift difference for similar parameters in adjacent sleep cycles, analyze the time offset and amplitude offset of parameter changes, determine the continuity of parameter drift direction, and generate a drift direction indication set.

[0008] S3: Based on the drift direction indication set, dynamically adjust the monitoring interval boundary of all parameters. When the drift direction is continuous and the amplitude offset increases, the monitoring range is proportionally enlarged. If the opposite is true, the monitoring range is proportionally shrunk. The monitoring interval is reconstructed with the current parameter mean as the center to generate a dynamic health baseline model.

[0009] S4: Based on the dynamic health baseline model, extract the risk coefficients of multiple parameters and calculate the risk contribution value by weighting. Normalize the risk contribution value of the same period and perform weighted fusion analysis of the risk index. Divide the risk level according to the risk index and generate personalized sleep health risk assessment results.

[0010] S5: Based on the personalized sleep health risk assessment results, perform trend fitting on the risk index of continuous cycles, predict the rate of change and trend direction of future cycle risks, and generate optimized personalized sleep health risk assessment results.

[0011] As a further aspect of the present invention, the multi-dimensional sleep key feature parameter set includes mean heart rate features, respiratory rate features, nighttime body movement features, skin temperature features, and environmental temperature and humidity features; the drift direction indicator set includes time offset direction indicator, amplitude offset direction indicator, and continuity state indicator; the dynamic health baseline model includes parameter mean distribution, upper and lower boundary distribution of monitoring interval, and parameter fluctuation amplitude distribution; the personalized sleep health risk assessment results include a risk coefficient set, a risk contribution value set, a risk index, and a risk level; and the optimized family sleep health risk results include the future cycle risk change rate, the future cycle risk trend direction, and the predicted cycle risk index.

[0012] As a further aspect of the present invention, the specific steps of S1 are as follows:

[0013] S101: Collects raw data streams of an individual's average heart rate, respiratory rate, number of nighttime body movements, skin temperature changes, bedroom temperature and humidity, noise intensity and air quality during the sleep cycle through a 5G edge computing terminal, and performs time synchronization and normalization to generate a time-series normalized parameter set.

[0014] S102: Based on the time-series normalized parameter set, calculate the fluctuations and correlations between multiple parameter dimensions, extract periodic features from the mean heart rate and respiratory rate, calculate the feature contribution of nighttime body movement frequency, skin temperature changes and bedroom environment, and filter key parameter dimensions by combining contribution thresholds to generate a feature component set;

[0015] S103: Based on the feature component set, perform weighted aggregation operation on the same parameter group, calculate the principal component ratio of multiple feature components in the overall feature space, establish a feature index sequence based on the ratio result, perform encoding integration on all index sequences, and generate a multi-dimensional sleep key feature parameter set.

[0016] As a further aspect of the present invention, the contribution threshold is obtained by calculating the correlation coefficient and variance contribution rate of multiple parameters among mean heart rate, respiratory rate, number of nighttime body movements, skin temperature changes and environmental factors, multiplying the correlation coefficient by the variance contribution rate to obtain the contribution value, and then analyzing the average value of the contribution of all parameters.

[0017] As a further aspect of the present invention, the specific steps of S2 are as follows:

[0018] S201: Based on the multi-dimensional sleep key feature parameter set, the time series data of the same type of parameter in adjacent sleep cycles are paired according to time, the difference value sequence of multiple parameters in adjacent cycles is calculated, and the positive and negative change intervals are identified for separation and labeling. The change amplitude and direction information of the difference value are analyzed to generate dynamic change trend.

[0019] S202: Invoke the dynamic change trend, perform joint calculation of time offset and amplitude offset on multiple parameter dimensions, perform normalization conversion on the two types of sequences, analyze the relative change trend of multiple parameters in the offset feature information, and generate parameter offset feature set;

[0020] S203: Based on the parameter offset feature set, perform continuity determination on the offset direction sequence of multiple parameters in continuous sleep cycles, compare the continuity of offset direction signs between adjacent cycles, remove parameters with discontinuous directions, summarize all direction identifiers and encode and integrate them to generate a drift direction indicator set.

[0021] The time offset is determined by calculating the cross-correlation function of two adjacent period parameter normalized time series, and is defined as the time delay point when the cross-correlation function reaches its maximum value;

[0022] The amplitude offset is defined as the arithmetic mean of the absolute values ​​of all elements in the difference value sequence within a complete sleep cycle, used to determine the continuity of the parameter drift direction.

[0023] As a further aspect of the present invention, the specific steps of S3 are as follows:

[0024] S301: Based on the drift direction indication set, read the monitoring interval boundaries of all parameters, extract the time series values ​​corresponding to multiple parameters, determine whether the direction signs are consistent for the drift direction sequence of each parameter, calculate the difference value of amplitude offset within adjacent periods, and generate an offset continuity identifier set.

[0025] S302: Based on the offset continuity identifier set, if the drift direction is detected to be continuous and the amplitude offset is increasing, the upper and lower limits of the interval are expanded according to the proportional amplification factor; if the offset is decreasing, it is contracted proportionally. The adjusted boundary spacing is calculated and recorded to generate the monitoring interval adjustment boundary set.

[0026] S303: Adjust the boundary set according to the monitoring interval, extract the mean value of the parameter sequence within the adjusted interval, construct a symmetrical interval range with the mean value as the center, call the adjusted boundary data to reposition the upper and lower limit positions of multiple parameters, and generate a dynamic health baseline model.

[0027] The rule for constructing the symmetrical interval is as follows: with the mean as the center, the upper and lower limits are expanded or contracted by 2 standard deviations respectively.

[0028] As a further aspect of the present invention, the specific steps of S4 are as follows:

[0029] S401: Based on the dynamic health baseline model, extract the numerical vectors of multiple parameters in the same period, calculate the deviation rate between the multiple parameters and the distribution mean, determine the risk weight of the parameters according to the proportion of deviation rate exceeding the risk coefficient threshold, record the risk weighting coefficients corresponding to the multiple parameters, and generate a set of multi-parameter risk coefficients.

[0030] S402: Call the multi-parameter risk coefficient set, perform a weighted product operation on the risk coefficient and weight value corresponding to each parameter, calculate the risk contribution value sequence of the multi-parameter, and then perform vector normalization on all risk contribution values ​​in the same period to generate a normalized risk contribution set;

[0031] S403: Based on the normalized risk contribution set, perform weighted fusion operation on the risk vectors of all parameters to calculate the overall risk index, and divide the risk index into intervals according to the risk level threshold, label the corresponding level, and generate personalized sleep health risk assessment results.

[0032] The risk coefficient threshold is set by statistical feature analysis of multi-parameter deviation rate samples in the dynamic health baseline model.

[0033] As a further aspect of the present invention, the risk level threshold is calculated by taking the mean and standard deviation of a multi-sample risk index, based on the concentration range and dispersion of the index distribution.

[0034] As a further aspect of the present invention, the specific steps of S5 are as follows:

[0035] S501: Based on the personalized sleep health risk assessment results, extract the risk index sequence of continuous periods, calculate the difference between the risk index values ​​of adjacent periods, extract the risk change rate parameter according to the change direction and magnitude of the difference sequence, and generate a set of periodic risk change rates.

[0036] S502: Call the set of periodic risk change rates, perform trend fitting operation on the multi-period risk change rates and corresponding risk indices, perform residual matching on the change rates and risk indices, determine the direction of risk change based on the sign of the slope of the fitted curve, and generate risk trend fitting results.

[0037] S503: Based on the risk trend fitting results, predict and calculate the risk index for future periods, calculate the rate of change of the risk index for multiple prediction periods and weight and fuse it with the original risk index, update the periodic risk index, and generate an optimized personalized sleep health risk assessment result.

[0038] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0039] This invention collects multiple physiological and environmental data points from an individual during their sleep cycle and performs time synchronization and normalization processing, resulting in more refined and accurate data processing. This process ensures consistency between different types of data, reduces errors caused by time deviations, and thus guarantees the effectiveness of subsequent analysis. Based on a set of key feature parameters extracted from multiple perspectives, the periodic drift difference analysis of similar parameters can promptly identify potential trends in health indicators, especially subtle changes that occur during sleep. This method not only monitors health risks more accurately but also dynamically adjusts the monitoring interval to respond to minor fluctuations in physiological state. Furthermore, by judging the continuity and magnitude of parameter changes, the monitoring range is adjusted in real time, effectively avoiding the risk of missing important data in traditional methods with fixed monitoring intervals, thus improving the flexibility and adaptability of monitoring. In addition, by generating a dynamic health baseline model and weighting the risk contribution value, an individual's sleep health status can be more comprehensively reflected, and potential risks can be assessed more precisely. This refined risk analysis not only allows for timely adjustments to health management strategies based on changes in risk indices but also provides personalized health warnings, thereby preventing potential health problems in advance. Overall, this innovation improves the accuracy and real-time nature of health monitoring by dynamically adjusting monitoring intervals and refining risk assessments, effectively avoiding the limitations of traditional static assessment methods. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a schematic diagram of the steps of the present invention; Detailed Implementation

[0042] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0043] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0044] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0045] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0046] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0047] Please see Figure 1 This invention provides a personalized sleep health risk assessment method based on 5G edge computing, comprising the following steps:

[0048] S1: Collect the average heart rate, respiratory rate, number of nighttime body movements, skin temperature changes, bedroom temperature and humidity, noise intensity and air quality of an individual during the sleep cycle through a 5G edge computing terminal, and perform time synchronization and normalization, extract key feature components, and generate a multi-dimensional set of key sleep feature parameters.

[0049] S2: Based on the key feature parameter set, calculate the cycle drift difference for similar parameters in adjacent sleep cycles, analyze the time offset and amplitude offset of parameter changes, determine the continuity of parameter drift direction, and generate a drift direction indication set.

[0050] S3: Based on the drift direction indicator set, the monitoring interval boundary of all parameters is dynamically adjusted. When the drift direction is continuous and the amplitude offset increases, the monitoring range is proportionally enlarged. If the opposite is true, the monitoring range is proportionally shrunk. The monitoring interval is reconstructed with the current parameter mean as the center to generate a dynamic health baseline model.

[0051] S4: Based on the dynamic health baseline model, extract the risk coefficients of multiple parameters and calculate the risk contribution value by weighting. Normalize the risk contribution value of the same period and perform weighted fusion analysis of the risk index. Divide the risk level according to the risk index and generate personalized sleep health risk assessment results.

[0052] S5: Based on the results of personalized sleep health risk assessment, the risk index of continuous cycles is fitted with trends to predict the rate of change and trend direction of future cycle risks, and to generate optimized personalized sleep health risk assessment results.

[0053] The key feature parameter set includes mean heart rate, respiratory rate, nighttime body movement, skin temperature, and ambient temperature and humidity. The drift direction indicator set includes time offset direction indicator, amplitude offset direction indicator, and continuity state indicator. The dynamic health baseline model includes parameter mean distribution, upper and lower boundary distribution of monitoring interval, and parameter fluctuation amplitude distribution. The personalized sleep health risk assessment results include risk coefficient set, risk contribution value set, risk index, and risk level. The optimized family sleep health risk results include future cycle risk change rate, future cycle risk trend direction, and predicted cycle risk index.

[0054] The specific steps of S1 are as follows:

[0055] S101: Collects raw data streams of an individual's average heart rate, respiratory rate, number of nighttime body movements, skin temperature changes, bedroom temperature and humidity, noise intensity and air quality during the sleep cycle through a 5G edge computing terminal, and performs time synchronization and normalization to generate a time-series normalized parameter set.

[0056] First, non-contact 5G millimeter-wave radar sensors, infrared temperature sensors, and environmental monitoring sensors deployed in the bedroom continuously collect raw data from an individual during an 8-hour sleep cycle from 22:00 to 06:00 the next day at a frequency of 1Hz. Specifically, the millimeter-wave radar sensor collects vital sign data, including heart rate (bpm) and respiratory rate (breaths / min); by processing the amplitude and phase changes of the radar signal, it simultaneously obtains the number of nighttime body movements (times); the infrared temperature sensor monitors and records changes in skin temperature at the individual's wrist (degrees Celsius); and the environmental monitoring sensor simultaneously collects environmental parameters in the bedroom, specifically temperature (°C), humidity (%RH), noise level (dB), and PM2.5 air quality concentration (μg / m³). The acquired heterogeneous data streams are transmitted to an edge computing gateway via a 5G network. At the gateway, the timestamps of all data streams are aligned using the Network Time Protocol (NTP) to ensure that the data from each sampling point corresponds to the same moment with millisecond precision. For example, at "T0+10 seconds," the collected data shows a heart rate of 62 bpm, a respiratory rate of 15 breaths / minute, 1 body movement within that minute, a skin temperature of 36.2°C, a bedroom temperature of 21.5°C, a humidity of 58%, a noise level of 34 dB, and an air quality of 12 μg / m³. Subsequently, the time-synchronized data is normalized. This normalization operation uses a min-max normalization method, linearly mapping the raw values ​​of each parameter to intervals. The maximum and minimum values ​​for each parameter are based on the results of a statistical experiment covering 500 healthy adults and continuously monitoring their sleep data for 90 days. This experiment determined the 1st and 99th percentiles of each parameter during sleep as the minimum and maximum boundaries for normalization. For example, the minimum boundary for heart rate is set at 45 bpm, and the maximum boundary at 95 bpm. For the previously collected heart rate value of 62 bpm, its normalized value is: (62-45) / (95-45) = 17 / 50 = 0.34. Similarly, if the minimum boundary of noise intensity is 25 dB and the maximum boundary is 75 dB, then the normalized value of noise intensity of 34 dB is: (34-25) / (75-25) = 9 / 50 = 0.18. This normalization calculation is performed on all seven parameter dimensions at all time points, resulting in a multi-dimensional time-series normalized parameter set.

[0057] S102: Based on the time-series normalized parameter set, calculate the fluctuation and correlation between multiple parameter dimensions, extract periodic features from mean heart rate and respiratory rate, calculate the feature contribution of nighttime body movement, skin temperature change and bedroom environment, and filter key parameter dimensions by combining contribution thresholds to generate feature component set;

[0058] The system receives the generated time-series normalized parameter set and performs calculations based on this dataset within consecutive 30-minute time windows. First, it extracts the periodic features of the mean heart rate and respiratory rate. This is achieved by calculating the autocorrelation function (ACF) of the normalized time series and determining the principal period by finding the time delay corresponding to the peak value of the ACF. For example, within a 30-minute window, if the normalized heart rate sequence shows a peak of 0.8 at a time delay of 60 seconds, its periodic feature is recorded as 60 seconds. Similarly, if the normalized respiratory rate sequence shows a peak of 0.9 at a time delay of 4 seconds, its periodic feature is recorded as 4 seconds. Next, it calculates the variability and correlation between multiple parameter dimensions. Variation is measured by the standard deviation of the normalized parameter values ​​within the 30-minute time window. Correlation is determined by calculating the Pearson correlation coefficient between any two parameter dimension time series, such as calculating the correlation coefficient between the mean heart rate sequence and the skin temperature change sequence. For five parameters—nighttime body movement frequency, skin temperature change, bedroom temperature and humidity, noise intensity, and air quality—the characteristic contribution was calculated. The contribution calculation benchmark came from a validation experiment involving 100 cases of polysomnography (PSG), which used the sleep fragmentation index determined by PSG as the gold standard for assessing sleep quality. Correlation analysis was performed between each parameter collected using this method and the corresponding time period's sleep fragmentation index (SFI), and the absolute value of the correlation coefficient was taken as the characteristic contribution of that parameter. For example, the calculated correlation coefficients between nighttime body movement frequency and SFI were 0.82, skin temperature change was -0.75, noise intensity was 0.61, and air quality was 0.45. Therefore, their contributions were 0.82, 0.75, 0.61, and 0.45, respectively. Subsequently, key parameter dimensions were selected based on contribution thresholds. The contribution thresholds were set based on a receiver operating characteristic (ROC) curve analysis experiment, which aimed to distinguish between stable and disordered sleep states, selecting the contribution value corresponding to the highest Youden index as the threshold. Analysis of 200 sets of experimental data determined the threshold to be 0.70. The contribution of each parameter was compared to this threshold: nighttime body movement frequency (0.82) was greater than 0.70 and was selected as a key parameter; skin temperature change (0.75) was greater than 0.70 and was selected as a key parameter; noise intensity (0.61) was less than 0.70 and was excluded. All five parameters were screened using this method, and the selected parameters, along with the previously extracted periodic features of heart rate and respiratory rate, constituted a feature component set.

[0059] S103: Based on the feature component set, perform weighted aggregation operation on the same parameter group, calculate the proportion of principal components of multiple feature components in the overall feature space, establish a feature index sequence based on the proportion result, perform encoding integration on all index sequences, and generate a multi-dimensional sleep key feature parameter set.

[0060] Based on the output feature component set, a weighted aggregation operation is first performed on the same type of parameter groups within the set. In this embodiment, the feature component set includes heart rate periodicity features, respiratory rate periodicity features, nighttime body movement frequency, and skin temperature changes. Heart rate and respiratory rate are divided into "physiological rhythm parameter groups," and nighttime body movement frequency and skin temperature changes are divided into "physiological state parameter groups." The weight of each parameter within the group is set according to its feature contribution calculated in S102. The setting process is as follows: the contribution of all parameters within the group is added to obtain the total contribution, and the weight of a single parameter is the ratio of its own contribution to the total contribution. The contribution of the heart rate periodicity feature is set to 0.88, and the contribution of the respiratory rate periodicity feature is set to 0.80. Then the total contribution of the "physiological rhythm parameter group" is 0.88 + 0.80 = 1.68. The weight of heart rate is 0.88 / 1.68 ≈ 0.524, and the weight of respiratory rate is 0.80 / 1.68 ≈ 0.476. If the quantized value of the heart rate periodicity feature is 85 and the respiratory rate is 92, then the aggregated feature components of this group are (85×0.524)+(92×0.476)=44.54+43.792=88.332. This method is used to calculate the aggregated feature components of all parameter groups. Subsequently, based on these aggregated feature components, their principal component proportions in the overall feature space are calculated. This process first constructs a data matrix composed of the aggregated feature components, then calculates the covariance matrix of this matrix, and performs eigenvalue decomposition on the covariance matrix to obtain a set of eigenvalues. The magnitude of each eigenvalue represents the amount of data variance explained by the corresponding principal component. The principal component proportion is the sum of all eigenvalues ​​divided by a single eigenvalue. For example, if two principal components are calculated, with corresponding eigenvalues ​​λ1=3.45 and λ2=1.05, then the sum of the eigenvalues ​​is 3.45+1.05=4.50. The first principal component has a proportion of 3.45 / 4.50 = 0.767, and the second principal component has a proportion of 1.05 / 4.50 = 0.233. Based on the principal component proportions from highest to lowest, a feature index sequence is established; in this example, the sequence is [1, 2]. Finally, all index sequences are encoded and integrated to generate a multi-dimensional set of key sleep feature parameters. The encoding rules are defined based on a preset proportion range. This rule was determined through a clustering analysis experiment involving 1000 samples. The experiment clustered the correlation between principal components with different proportion ranges and sleep stages (such as light sleep, deep sleep, and REM sleep) to establish the encoding system. For example, a proportion higher than 0.70 is defined as a "core component," encoded as 11; a proportion between 0.20 and 0.70 is defined as an "important component," encoded as 10; and a proportion lower than 0.20 is defined as a "minor component," encoded as 01. According to this rule, the proportion [0.767, 0.233] corresponding to the index sequence is encoded as [11, 10].The integrated encoding involves concatenating strings according to the index sequence to generate a multi-dimensional set of key sleep feature parameters: "1110".

[0061] Table 1. Example of raw data and normalized parameters for sleep cycles.

[0062]

[0063] As shown in Table 1, this table lists the raw physiological and environmental data collected at a specific time point and displays the results after maximum and minimum normalization calculations based on preset parameter ranges. These normalized values ​​form the input basis for subsequent analysis steps.

[0064] The specific steps of S2 are as follows:

[0065] S201: Based on the key feature parameter set, time series data of the same type of parameter in adjacent sleep cycles are paired according to time, the difference value sequence of multiple parameters in adjacent cycles is calculated, and positive and negative change intervals are identified for separation and labeling. The magnitude and direction information of the difference value changes are analyzed to generate dynamic change trends.

[0066] Based on the generated set of key feature parameters, the parameters corresponding to the core component coded as "11"—namely, mean heart rate and number of nighttime body movements—are selected for periodic drift analysis. First, these two parameters are paired one-to-one by timestamp using the time-series normalized parameter sets generated by S101 across two consecutive sleep cycles (defined as cycle one and cycle two, e.g., Monday night and Tuesday night sleep). After pairing, the difference in the normalized parameter values ​​at each time point in adjacent cycles is calculated, generating a sequence of difference values. Taking a 5-minute segment with a sampling frequency of 1 time / minute as an example, if the normalized heart rate value sequence for period one is [0.34, 0.36, 0.35, 0.38, 0.37], and the corresponding sequence for period two is [0.38, 0.39, 0.36, 0.41, 0.40], then by subtracting the values ​​of period two from the values ​​of period one, the difference value sequence is calculated as: [0.04, 0.03, 0.01, 0.03, 0.03]. Similarly, if the normalized sequence of nighttime body movement frequency is [0.10, 0.20, 0.10, 0.30, 0.10] in period one and [0.10, 0.15, 0.20, 0.25, 0.15] in period two, then its difference value sequence is: [0.00, -0.05, 0.10, -0.05, 0.05]. Subsequently, the positive and negative variation intervals of this difference value sequence are identified and labeled. Specifically, each value in the difference value sequence is iterated; if the value is greater than 0, it is labeled "+1"; if the value is less than 0, it is labeled "-1"; if the value is equal to 0, it is labeled "0". The difference value sequence of heart rate is all positive, and its direction of change is labeled as [+1, +1, +1, +1, +1]. The corresponding direction of change for the difference value sequence of nighttime body movement frequency is labeled as [0, -1, +1, -1, +1]. Next, the magnitude and direction of the difference values ​​are analyzed; the magnitude is the absolute value of the difference. For example, the magnitude of the change in the number of nighttime body movements at the second time point is |-0.05|=0.05, and the direction is negative. The calculated difference value sequences, change direction indicator sequences, and magnitude information of each parameter are compiled to generate a dynamic trend.

[0067] S202: Invoke the dynamic change trend, perform joint calculation of time offset and amplitude offset on multiple parameter dimensions, perform normalization conversion on the two types of sequences, analyze the relative change trend of multiple parameters in the offset feature information, and generate parameter offset feature set;

[0068] The generated dynamic trend is invoked to perform joint calculations of time offset and amplitude offset for the two parameters: mean heart rate and number of nighttime body movements. First, the amplitude offset is calculated, defined as the arithmetic mean of the absolute values ​​of all elements in the difference sequence within a complete sleep cycle (8 hours). This calculation is based on a 14-day follow-up experiment of sleep data from 200 healthy individuals. The results show that the mean amplitude offset stably reflects the overall fluctuation of the parameter between adjacent sleep days. Taking heart rate as an example, if the sum of the absolute values ​​of the difference sequence within a complete cycle is 134.4, and the total number of sampling points is 28800, then its amplitude offset is 134.4 / 28800 = 0.00467. Next, the time offset is calculated. This value is determined by calculating the cross-correlation function of the normalized time series of two adjacent cycles. The time offset is defined as the time delay point at which the cross-correlation function reaches its maximum value. For example, calculating the cross-correlation between the normalized heart rate sequences of cycle one and cycle two reveals that the cross-correlation coefficient reaches a peak of 0.85 at a time delay of +120 seconds; therefore, the time offset of heart rate is recorded as +120 seconds. Subsequently, normalization is performed on the two types of offsets. The normalization range is set based on a validation experiment containing 500 sets of data. This experiment statistically analyzed the distribution of time and amplitude offsets in individuals with different degrees of sleep disturbance, and used the 5th and 95th percentiles as the normalization boundaries. The normalization range for amplitude offset is set to [0, 0.01], and the range for time offset is [-600 seconds, +600 seconds]. The normalized value for the heart rate amplitude offset of 0.00467 is (0.00467-0) / (0.01-0) = 0.467. The normalized value of its time offset + 120 seconds is 120 - (-600) / 600 - (-600) = 720 / 1200 = 0.60. After performing this calculation on all key parameters, the relative change trends of multiple parameters in the offset feature information are analyzed. By comparing the normalized offsets of each parameter, their relative stability is determined. For example, if the normalized amplitude offset of the number of nighttime body movements is 0.75, which is higher than that of the heart rate (0.467), it indicates that the number of nighttime body movements fluctuates more between two cycles. The normalized time offsets and amplitude offsets of all parameters are combined to generate a parameter offset feature set.

[0069] S203: Based on the parameter offset feature set, perform continuity determination on the offset direction sequence of multiple parameters in consecutive sleep cycles, compare the continuity of offset direction signs between adjacent cycles, remove parameters with discontinuous directions, summarize all direction identifiers and encode and integrate them to generate a drift direction indicator set.

[0070] The time offset is determined by calculating the cross-correlation function of two adjacent periodic parameter normalized time series, and is defined as the time delay point when the cross-correlation function reaches its maximum value;

[0071] The amplitude offset is defined as the arithmetic mean of the absolute values ​​of all elements in the difference value sequence within a complete sleep cycle, and is used to determine the continuity of the parameter drift direction.

[0072] Based on the generated parameter offset feature set, the continuity of the offset direction sequence of the two parameters, mean heart rate and nocturnal body movements, is determined across three consecutive sleep cycles (cycle one, cycle two, and cycle three). This process requires calculating the dynamic trends of cycle two relative to cycle one (denoted as drift 1-2) and cycle three relative to cycle two (denoted as drift 2-3) using the S201 method. From these two difference sets, the dominant change direction sign of each parameter at the vast majority of time points (over 85%) is extracted. This 85% threshold was determined through an experiment analyzing multi-day data from 100 stable sleepers, finding that the consistency of diurnal variation direction of their key physiological parameters was generally higher than this value. For example, for heart rate, the dominant direction of drift 1-2 is "+", and the dominant direction of drift 2-3 is also "+". For nocturnal body movements, the dominant direction of drift 1-2 is "-", while the dominant direction of drift 2-3 becomes "+". Subsequently, the continuity of the offset direction sign between adjacent cycles (i.e., drift 1-2 and drift 2-3) is compared. The direction sign for heart rate changed from "+" to "+", indicating a continuous direction. The direction sign for nighttime body movement count changed from "-" to "+", indicating a discontinuous direction. Next, parameters with discontinuous directions were removed; therefore, nighttime body movement count was removed from the current analysis sequence. After determining and filtering all key parameters, the direction signs of all retained parameters were summarized. In this example, only heart rate was retained, with its direction sign "+" in drift 2-3. Finally, the summarized direction signs were encoded and integrated. The encoding rule was pre-set: "+" was encoded as "1", "0" as "0", and "-" as "-1". This rule was established to simplify subsequent data processing. Therefore, the "+" direction for heart rate was encoded as "1". If other parameters (such as respiratory rate) were also retained with a "-" direction, they were encoded as "-1", resulting in a final integration of "1, -1". In this embodiment, the final generated drift direction indicator set is "1".

[0073] Table 2 Example of Periodic Drift Difference Calculation

[0074]

[0075] As shown in Table 2, this table presents normalized data collected from two adjacent sleep cycles at three consecutive time points for two parameters: heart rate and the number of nighttime body movements. The differences between these parameters and their corresponding directional indicators are calculated. This result constitutes the core content of the dynamic trend.

[0076] The specific steps for S3 are as follows:

[0077] S301: Based on the drift direction indication set, read the monitoring interval boundaries of all parameters, extract the time series values ​​corresponding to multiple parameters, determine whether the direction signs are consistent for the drift direction sequence of each parameter, calculate the difference value of amplitude offset within adjacent periods, and generate an offset continuity identifier set.

[0078] Based on the generated drift direction indicator set, which includes a drift direction code of "1" for the heart rate parameter (representing a positive drift), the initial monitoring interval boundary for this parameter is first read. This boundary is defined as [45, 95] in S101, with units of beats per minute (bpm). Next, the normalized time series values ​​of the heart rate parameter, aligned to timestamps, are extracted from three consecutive sleep cycles (cycle one, cycle two, and cycle three). For each parameter's drift direction sequence, it is determined whether the direction sign remains consistent. This determination is made by comparing the dominant direction signs of the two consecutive drift stages: "cycle two relative to cycle one" (drift 1-2) and "cycle three relative to cycle two" (drift 2-3). As shown in S203, the dominant direction of the heart rate in the drift 1-2 stage is positive, and its drift direction indicator set code is "1". To determine continuity, the direction of the drift 2-3 stage needs to be calculated. By calculating the difference sequence of normalized heart rate values ​​between cycle three and cycle two, it was found that the difference was positive at more than 85% of the time points, therefore the main direction of drift phase 2-3 was also positive. Since the direction signs of the two consecutive phases were both positive, it was determined that the drift direction of the heart rate parameter was consistent. Subsequently, the difference value of amplitude shift within adjacent cycles was calculated. According to the calculation in S202, the amplitude shift of heart rate in drift phase 1-2 was 0.00467. The amplitude shift in drift phase 2-3 was calculated using the same method: first, the difference sequence of normalized heart rate values ​​between cycle three and cycle two was obtained, and the sum of the absolute values ​​of all elements in this sequence was calculated and set to 158.4; then, this sum was divided by the total number of sampling points, 28800, to obtain the amplitude shift in drift phase 2-3 as 158.4 / 28800 = 0.0055. Finally, the difference value is calculated, which is the amplitude offset of drift 2-3 minus the amplitude offset of drift 1-2: 0.0055 - 0.00467 = 0.00083. This positive value indicates that the amplitude offset is increasing. The direction continuity determination result (yes) is combined with the amplitude offset difference value (+0.00083) to generate an offset continuity identifier for the heart rate parameter, which constitutes the offset continuity identifier set.

[0079] S302: Based on the offset continuity identifier set, if the drift direction is continuous and the amplitude offset is increasing, the upper and lower limits of the interval are expanded according to the proportional amplification factor. If the offset is decreasing, it is contracted proportionally. The adjusted boundary spacing is calculated and recorded to generate the monitoring interval adjustment boundary set.

[0080] Based on the generated set of offset continuity identifiers, heart rate parameters identified as having continuous direction are processed. First, the trend of amplitude offset is detected. Since the amplitude offset difference calculated by S301 is +0.00083, which is greater than 0, it is determined to be an increasing trend. According to preset rules, when continuous drift direction and increasing amplitude offset are detected, the monitoring interval boundary of this parameter is expanded. This expansion operation is accomplished through a scaling factor. This scaling factor is set to 1.5. This value is based on a 60-day experimental study that monitored continuous sleep data of 250 participants. By minimizing the root mean square error between the predicted boundary and the 99th percentile of the actual data the next day, 1.5 was verified and determined to be the optimal scaling factor. The specific expansion calculation process is as follows: First, the normalized amplitude offset difference value of 0.00083 is converted to the original unit (bpm). This conversion utilizes the normalized range width of the S101 central rate (95 bpm - 45 bpm = 50 bpm), yielding an actual fluctuation change of 0.00083 × 50 = 0.0415 bpm. Next, this change is multiplied by a scaling factor of 1.5 to obtain the total boundary adjustment width: 0.0415 bpm × 1.5 = 0.06225 bpm. This width is evenly distributed across the upper and lower limits of the interval. The upper limit is expanded by 0.06225 / 2 = 0.031125 bpm, and the lower limit is also contracted by 0.031125 bpm. This adjustment value is applied to the initial monitoring interval [45, 95] bpm. The adjusted new lower limit is 45 - 0.031125 = 44.968875 bpm; the new upper limit is 95 + 0.031125 = 95.031125 bpm. If the amplitude offset shows a decreasing trend (negative difference value), then proportional contraction is performed. The adjusted boundary spacing is calculated and recorded; the new spacing is 95.031125 - 44.968875 = 50.06225 bpm. The heart rate parameter and its adjusted boundary [44.968875, 95.031125] are recorded as a data pair to generate the monitoring interval adjustment boundary set.

[0081] S303: Adjust the boundary set according to the monitoring interval, extract the mean of the parameter sequence within the adjusted interval, construct a symmetrical interval range with the mean as the center, call the adjusted boundary data to reposition the upper and lower limits of multiple parameters, and generate a dynamic health baseline model.

[0082] The rule for constructing symmetrical intervals is: with the mean as the center, the upper and lower limits are expanded or contracted by 2 standard deviations respectively;

[0083] Based on the generated monitoring interval adjustment boundary set, the adjusted monitoring interval for the heart rate parameter [44.968875, 95.031125] is extracted. Next, all time-series values ​​of this parameter in the next sleep cycle (i.e., cycle three) that fall within the adjusted interval are extracted. The arithmetic mean of these valid time-series values ​​is calculated. For example, in the 8-hour sleep cycle three, a total of 28,800 heart rate data points were collected, of which 28,750 data points were located within the interval [44.968875, 95.031125], with a sum of 1,901,400. The calculated sequence mean is 1901,400 / 28,750 = 66.135 bpm. A symmetrical steady-state interval is constructed centered on this mean of 66.135 bpm. The width of this symmetrical interval is determined based on the standard deviation of the valid data sequence in cycle three. First, the standard deviation of this sequence is calculated, assuming a result of 4.5 bpm. The steady-state interval was constructed using the rule: mean ± 2 standard deviations. This rule was established based on a statistical analysis of 10,000 sleep records, which showed that the interval defined in this way could contain 95.4% of the data points of an individual during non-wakeful sleep. According to this rule, the upper and lower limits of the symmetrical interval were calculated: the upper limit was 66.135 + (2 × 4.5) = 75.135 bpm; the lower limit was 66.135 - (2 × 4.5) = 57.135 bpm. The resulting steady-state parameter distribution range was [57.135, 75.135] bpm. Subsequently, the adjusted boundary data [44.968875, 95.031125] generated by S302 and the newly calculated steady-state distribution interval [57.135, 75.135] bpm were used to reposition the monitoring range of the heart rate parameter. This positioning process uses the newly generated steady-state distribution range as the individual's current core stable range, while the adjusted monitoring boundary serves as the allowable external boundary for fluctuations. The parameter names, the new core stable range, and the external boundary are combined to generate a dynamic health baseline model.

[0084] Table 3. Examples of Parameter Interval Dynamic Adjustment and Steady-State Distribution Generation

[0085]

[0086] As shown in Table 3, this table clearly demonstrates how the monitoring intervals for heart rate parameters are dynamically adjusted based on continuous periodic drift, and generates a core distribution interval that better reflects the current physiological homeostasis of an individual based on the data characteristics of the latest sleep cycle. All values ​​are rounded to two decimal places for display.

[0087] The specific steps of S4 are as follows:

[0088] S401: Based on the dynamic health baseline model, extract the numerical vectors of multiple parameters in the same period, calculate the deviation rate between the multiple parameters and the distribution mean, determine the parameter risk weights according to the proportion of deviation rates exceeding the risk coefficient threshold, record the risk weighting coefficients corresponding to the multiple parameters, and generate a set of multi-parameter risk coefficients.

[0089] Based on the generated dynamic health baseline model, the numerical vectors of heart rate and respiratory rate, two parameters contained within it, were extracted over period three. Specifically, the steady-state distribution mean of heart rate (66.14 bpm) and the steady-state distribution mean of respiratory rate (15.5 breaths / minute) were extracted. Subsequently, for the complete 8-hour time series (28,800 sampling points) of these two parameters over period three, the deviation rate between the value of each sampling point and the mean of the corresponding parameter distribution was calculated. The deviation rate was calculated by dividing the absolute value of the difference between the instantaneous measurement and the mean by the mean. For example, at a certain moment, the heart rate measurement was 78 bpm, and its deviation rate was |78-66.14| / 66.14≈0.179. Then, whether the data point was considered a high deviation point was determined based on whether the deviation rate exceeded the risk coefficient threshold. The risk coefficient threshold was set based on a controlled experiment involving 500 subjects. This experiment correlated deviation rate data with micro-arousals identified through polysomnography (PSG). Using receiver operating characteristic (ROC) curve analysis, the point with the highest Youden index in distinguishing between stable and unstable sleep states was selected, and the threshold was determined to be 0.15. Therefore, the aforementioned deviation rate of 0.179 is greater than 0.15, and this data point is counted as a high deviation. After traversing the data throughout the entire cycle, the proportion of high deviation points to the total number of sampling points is calculated; this proportion is the parameter risk weight. If heart rate shows 4608 high deviation points in cycle three, its risk weight is 4608 / 28800 = 0.16. Similarly, if respiratory rate shows 2304 high deviation points, its risk weight is 2304 / 28800 = 0.08. Finally, the risk weighting coefficients corresponding to these two parameters are recorded. This coefficient is a preset value reflecting the prior importance of each physiological parameter to the overall sleep health risk. This coefficient was determined through a Delphi method survey of 30 sleep medicine experts. Based on clinical experience and existing research, the experts scored the correlation between each parameter and long-term cardiovascular and nervous system health risks on a 10-point scale, and the average score was used as the risk weighting coefficient. The risk weighting coefficient for heart rate was set at 8.5, and for respiratory rate at 7.5. The parameters, risk weighting coefficients, and calculated risk weights were combined to generate a multi-parameter risk coefficient set.

[0090] S402: Call the multi-parameter risk coefficient set, perform a weighted product operation on the risk coefficient and weight value corresponding to each parameter, calculate the risk contribution value sequence of the multi-parameter, and then perform vector normalization on all risk contribution values ​​in the same period to generate a normalized risk contribution set;

[0091] The generated multi-parameter risk coefficient set is invoked, containing two entries: heart rate (risk weighting coefficient 8.5, risk weight 0.16) and respiratory rate (risk weighting coefficient 7.5, risk weight 0.08). First, for each parameter, its corresponding risk weighting coefficient and risk weight value are multiplied by a weighted product to calculate the risk contribution value for each parameter. For heart rate, the risk contribution value is 8.5 × 0.16 = 1.36. For respiratory rate, the risk contribution value is 7.5 × 0.08 = 0.60. These two values ​​constitute the risk contribution value sequence [1.36, 0.60] within cycle three. Subsequently, vector normalization is performed on this risk contribution value sequence [1.36, 0.60]. This normalization uses the L2 norm (Euclidean norm) method, first calculating the magnitude of the vector, i.e., the square root of the sum of the squares of each component. The calculation process is as follows: The modulus of the vector = (1.36² + 0.60²)^(1 / 2) = (1.8496 + 0.36)^(1 / 2) = (2.2096)^(1 / 2) ≈ 1.4865. Then, each element in the risk contribution value sequence is divided by this modulus value to obtain the normalized risk contribution value. The normalized risk contribution value of heart rate is 1.36 / 1.4865 ≈ 0.9149. The normalized risk contribution value of respiratory rate is 0.60 / 1.4865 ≈ 0.4036. These two normalized values ​​together constitute the normalized risk contribution set. This set reflects the relative contribution of different physiological parameters to the overall risk within the current sleep cycle. The larger the value, the higher the relative contribution of abnormal fluctuations in that parameter to the overall risk. In this example, the relative contribution of heart rate (0.9149) is significantly higher than that of respiratory rate (0.4036). Record each parameter and its corresponding normalized risk contribution value to generate a normalized risk contribution set.

[0092] S403: Based on the normalized risk contribution set, perform weighted fusion calculation on the risk vectors of all parameters to calculate the overall risk index, and divide the risk index into intervals according to the risk level threshold, label the corresponding level, and generate personalized sleep health risk assessment results.

[0093] The risk coefficient threshold is set by statistical characteristic analysis of the multi-parameter deviation rate samples in the dynamic health baseline model;

[0094] Based on the generated normalized risk contribution set and the unnormalized risk contribution value sequence [1.36, 0.60] generated during the calculation process, a weighted fusion operation is performed on the risk contribution values ​​of all parameters to calculate the overall sleep risk index. In this embodiment, the weighted fusion operation is defined as directly summing the unnormalized risk contribution values ​​of each parameter, i.e., the fusion weight is 1 for each parameter. The calculation process is as follows: Overall risk index = heart rate risk contribution value + respiratory rate risk contribution value = 1.36 + 0.60 = 1.96. This index integrates the abnormal fluctuations of multiple key physiological parameters. Subsequently, the calculated overall risk index of 1.96 is divided into intervals according to a preset risk level threshold. The setting of this risk level threshold is based on a one-year follow-up study of 1000 individuals with different health conditions. This study statistically correlated the calculated overall risk index with the Pittsburgh Sleep Quality Index (PSQI) score and the subsequent incidence of cardiovascular events, and determined the level based on the 75th and 95th percentiles of the risk index distribution. The specific classification criteria are as follows: an overall risk index between 0.0 and 1.5 is defined as "low risk" and labeled "Level I"; an index between 1.5 and 3.0 is defined as "moderate risk" and labeled "Level II"; and an index greater than 3.0 is defined as "high risk" and labeled "Level III". The calculated overall risk index of 1.96 is compared with this standard: 1.5 < 1.96 ≤ 3.0, therefore the index falls within the "moderate risk" range. Finally, the corresponding risk level label "Level II" is assigned, and the index and level label are integrated to generate a personalized sleep health risk assessment result: "Overall risk index is 1.96, risk level is Level II (moderate risk)".

[0095] The specific steps of S5 are as follows:

[0096] S501: Based on the results of personalized sleep health risk assessment, extract the risk index sequence of continuous periods, calculate the difference between the risk index values ​​of adjacent periods, extract the risk change rate parameter according to the change direction and magnitude of the difference sequence, and generate a set of periodic risk change rates.

[0097] Based on the generated personalized sleep health risk assessment result, namely the overall risk index of 1.96 for cycle three, a risk index sequence of three consecutive sleep cycles, including this result, is first extracted. This sequence is obtained by continuously executing steps S401 to S403. For example, the overall risk index of cycle one is 1.80, the overall risk index of cycle two is 1.90, and together with the 1.96 of cycle three, they form a time series: [1.80, 1.90, 1.96]. Next, the difference between the risk index values ​​of these adjacent cycles is calculated to determine the change in risk. The difference in the risk index of cycle two relative to cycle one is calculated as 1.90 minus 1.80, resulting in 0.10. Similarly, the difference in the risk index of cycle three relative to cycle two is calculated as 1.96 minus 1.90, resulting in 0.06. These two difference values ​​constitute the difference sequence [0.10, 0.06]. Based on the direction and magnitude of change in this difference sequence, the risk change rate parameter is extracted. The direction of change is determined by the sign of the difference value. Since both 0.10 and 0.06 are positive, it indicates that the risk index shows an upward trend in both of these consecutive transitions. The magnitude of change is the absolute value of the difference value. The risk change rate parameter is defined here as the amount of change in the risk index within each period interval (one unit period). Therefore, the risk change rate of period two relative to period one is +0.10, and the risk change rate of period three relative to period two is +0.06. Each period interval and its corresponding risk change rate parameter are recorded as a data pair, for example, (period 1-2, +0.10) and (period 2-3, +0.06). These data pairs are aggregated to generate a set of periodic risk change rates.

[0098] S502: Call the set of periodic risk change rates, perform trend fitting operation on the multi-period risk change rates and corresponding risk indices, perform residual matching on the change rates and risk indices, determine the risk change direction based on the sign of the slope of the fitted curve, and generate risk trend fitting results.

[0099] The generated set of periodic risk change rates and the corresponding continuous risk index sequence [1.80, 1.90, 1.96] are used to perform trend fitting operations on the risk indices for these three periods. This operation uses the least squares method to linearly fit the changes in the risk indices over time (period numbers 1, 2, 3) to determine their long-term trend. A data point set is established, where the independent variable is the period number and the dependent variable is the corresponding risk index: (1, 1.80), (2, 1.90), (3, 1.96). First, the slope of the fitted straight line is calculated as: Risk Index = Slope × Period Number + Intercept. The slope is calculated as follows: Multiply the number of data points (3) by the sum of the products of the period number and the risk index (1×1.80+2×1.90+3×1.96=11.48), resulting in 34.44; then multiply the sum of the period numbers (1+2+3=6) by the sum of the risk indices (1.80+1.90+1.96=5.66), resulting in 33.96; subtract the two to get 0.48. Next, multiply the number of data points (3) by the sum of the squares of the period numbers (1²+2²+3²=14), resulting in 42; then subtract the square of the sum of the period numbers (6²=36), resulting in 6. Finally, 0.48 / 6=0.08. Next, residual matching was performed between the rate of change and the risk index. First, the intercept was calculated: subtracting the product of the slope 0.08 and the sum of the period numbers (6) (0.48) from the sum of the risk indices (5.66), we get 5.18; then, 5.18 / 3 = 1.727. The fitted equation is: Risk Index = 0.08 × Period Number + 1.727. The residual (actual value minus predicted value) for each point was calculated: the residual for period one is 1.80 - (0.08 × 1 + 1.727) = -0.007; the residual for period two is 1.90 - (0.08 × 2 + 1.727) = +0.013; the residual for period three is 1.96 - (0.08 × 3 + 1.727) = -0.007. Finally, the overall direction of risk change was determined based on the sign of the slope of the fitted curve. Since the slope 0.08 is positive, the direction of risk change was determined to be a continuous increase. The slope, intercept, residual sequence, and direction of change determination results obtained from the fitting are integrated to generate the risk trend fitting result.

[0100] S503: Based on the risk trend fitting results, predict and calculate the risk index for future cycles, calculate the rate of change of the risk index for multiple prediction cycles and weight and fuse it with the original risk index, update the periodic risk index, and generate optimized personalized sleep health risk assessment results.

[0101] Based on the generated risk trend fitting results, i.e., the fitted equation risk index = 0.08 × cycle number + 1.727 and the judgment of increasing risk, the risk index for the next cycle (cycle four) is predicted. Substituting cycle number 4 into the fitted equation, the predicted risk index is 0.08 × 4 + 1.727 = 2.047. Next, the rate of change of the predicted cycle risk index relative to the previous actual cycle is calculated. This rate of change is the difference between the predicted cycle four risk index of 2.047 and the actual cycle three risk index of 1.96, i.e., 2.047 - 1.96 = 0.087. Subsequently, this rate of change value is weighted and fused with the original latest risk index (cycle three 1.96) to update the cyclical risk index. The fusion weight here is set as the coefficient of determination for trend fitting, which reflects the goodness of fit. This setting is verified by a backtesting experiment on 200 sets of historical sequence data. The results show that using the coefficient of determination as the weight can reduce the average absolute error of the next cycle prediction by 18% compared with fixed weights. The coefficient of determination is calculated as follows: the sum of squares of the residuals is (-0.007)² + (0.013)² + (-0.007)² ≈ 0.000267; the total sum of squares is (1.80 - 1.887)² + (1.90 - 1.887)² + (1.96 - 1.887)² ≈ 0.01307; therefore, the coefficient of determination = 1 - (0.000267 / 0.01307) ≈ 0.9796. Applying this weight to the weighted fusion calculation: the updated risk index = cycle three risk index + coefficient of determination × predicted rate of change = 1.96 + 0.9796 × 0.087 ≈ 1.96 + 0.0852 = 2.0452. Finally, the updated periodic risk index of 2.0452 is rated according to the risk level threshold defined in S403. Since 1.5 < 2.0452 ≤ 3.0, its risk level remains "Level II (Moderate Risk)". This updated index and risk level are then integrated to generate an optimized personalized sleep health risk assessment result.

[0102] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A personalized sleep health risk assessment method based on 5G edge computing, characterized in that, Includes the following steps: S1: Collect physiological time series data and environmental time series data of individuals during their sleep cycle through 5G edge computing terminals, synchronize and normalize the data, extract feature indicators related to sleep health, and construct a multi-dimensional set of key sleep feature parameters. S2: Based on the multi-dimensional sleep key feature parameter set, calculate the dynamic change trend of the same feature indicators of adjacent sleep cycles, quantify the time offset and amplitude offset according to the dynamic change trend, determine the continuity of parameter drift direction, and generate a drift direction indication set. The time offset is determined by calculating the cross-correlation function of two adjacent period parameter normalized time series, and is defined as the time delay point when the cross-correlation function reaches its maximum value; The amplitude offset is defined as the arithmetic mean of the absolute values ​​of all elements in the difference value sequence within a complete sleep cycle, used to determine the continuity of the parameter drift direction; The specific steps for S3 are as follows: S301: Based on the drift direction indication set, read the monitoring interval boundaries of all parameters, extract the time series values ​​corresponding to multiple parameters, determine whether the direction signs are consistent for the drift direction sequence of each parameter, calculate the difference value of amplitude offset within adjacent periods, and generate an offset continuity identifier set. S302: Based on the offset continuity identifier set, if the drift direction is detected to be continuous and the amplitude offset is increasing, the upper and lower limits of the interval are expanded according to the proportional amplification factor; if the offset is decreasing, it is contracted proportionally. The adjusted boundary spacing is calculated and recorded to generate the monitoring interval adjustment boundary set. S303: Adjust the boundary set according to the monitoring interval, extract the mean value of the parameter sequence within the adjusted interval, construct a symmetrical interval range with the mean value as the center, call the adjusted boundary data to reposition the upper and lower limit positions of multiple parameters, and generate a dynamic health baseline model. The rule for constructing the symmetrical interval is: with the mean as the center, the upper and lower limits are expanded or contracted by 2 standard deviations respectively; S4: Based on the dynamic health baseline model, extract the risk coefficients of multiple parameters and calculate the risk contribution value by weighting. Normalize the risk contribution value of the same period and perform weighted fusion analysis of the risk index. Divide the risk level according to the risk index and generate personalized sleep health risk assessment results.

2. The personalized sleep health risk assessment method based on 5G edge computing according to claim 1, characterized in that, The multidimensional sleep key feature parameter set includes mean heart rate features, respiratory rate features, nighttime body movement features, skin temperature features, and environmental temperature and humidity features. The drift direction indicator set includes time offset direction indicator, amplitude offset direction indicator, and continuity state indicator. The dynamic health baseline model includes parameter mean distribution, upper and lower boundary distribution of monitoring interval, and parameter fluctuation amplitude distribution. The personalized sleep health risk assessment results include risk coefficient set, risk contribution value set, risk index, and risk level.

3. The personalized sleep health risk assessment method based on 5G edge computing according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Collects raw data streams of an individual's average heart rate, respiratory rate, number of nighttime body movements, skin temperature changes, bedroom temperature and humidity, noise intensity and air quality during the sleep cycle through a 5G edge computing terminal, and performs time synchronization and normalization to generate a time-series normalized parameter set. S102: Based on the time-series normalized parameter set, calculate the fluctuations and correlations between multiple parameter dimensions, extract periodic features from the mean heart rate and respiratory rate, calculate the feature contribution of nighttime body movement frequency, skin temperature changes and bedroom environment, and filter key parameter dimensions by combining contribution thresholds to generate a feature component set; S103: Based on the feature component set, perform weighted aggregation operation on the same parameter group, calculate the principal component ratio of multiple feature components in the overall feature space, establish a feature index sequence based on the ratio result, perform encoding integration on all index sequences, and generate a multi-dimensional sleep key feature parameter set.

4. The personalized sleep health risk assessment method based on 5G edge computing according to claim 3, characterized in that, The contribution threshold is calculated by multiplying the correlation coefficient and variance contribution rate of multiple parameters among mean heart rate, respiratory rate, number of nighttime body movements, skin temperature changes and environmental factors to obtain the contribution value, and then analyzing the average value of the contribution of all parameters.

5. The personalized sleep health risk assessment method based on 5G edge computing according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Based on the multi-dimensional sleep key feature parameter set, the time series data of the same type of parameter in adjacent sleep cycles are paired according to time, the difference value sequence of multiple parameters in adjacent cycles is calculated, and the positive and negative change intervals are identified for separation and labeling. The change amplitude and direction information of the difference value are analyzed to generate dynamic change trend. S202: Invoke the dynamic change trend, perform joint calculation of time offset and amplitude offset on multiple parameter dimensions, perform normalization conversion on the two types of sequences, analyze the relative change trend of multiple parameters in the offset feature information, and generate parameter offset feature set; S203: Based on the parameter offset feature set, perform continuity determination on the offset direction sequence of multiple parameters in continuous sleep cycles, compare the continuity of offset direction signs between adjacent cycles, remove parameters with discontinuous directions, summarize all direction identifiers and encode and integrate them to generate a drift direction indicator set.

6. The personalized sleep health risk assessment method based on 5G edge computing according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Based on the dynamic health baseline model, extract the numerical vectors of multiple parameters in the same period, calculate the deviation rate between the multiple parameters and the distribution mean, determine the risk weight of the parameters according to the proportion of deviation rate exceeding the risk coefficient threshold, record the risk weighting coefficients corresponding to the multiple parameters, and generate a set of multi-parameter risk coefficients. S402: Call the multi-parameter risk coefficient set, perform a weighted product operation on the risk coefficient and weight value corresponding to each parameter, calculate the risk contribution value sequence of the multi-parameter, and then perform vector normalization on all risk contribution values ​​in the same period to generate a normalized risk contribution set; S403: Based on the normalized risk contribution set, perform weighted fusion operation on the risk vectors of all parameters to calculate the overall risk index, and divide the risk index into intervals according to the risk level threshold, label the corresponding level, and generate personalized sleep health risk assessment results. The risk coefficient threshold is set by statistical feature analysis of multi-parameter deviation rate samples in the dynamic health baseline model.

7. A personalized sleep health risk assessment method based on 5G edge computing according to claim 6, characterized in that, The risk level threshold is calculated by taking the mean and standard deviation of a multi-sample risk index, based on the central tendency and dispersion of the index distribution.

8. The personalized sleep health risk assessment method based on 5G edge computing according to claim 1, characterized in that, The method further includes: S5: Based on the personalized sleep health risk assessment results, perform trend fitting on the risk index of continuous cycles, predict the rate of change and trend direction of future cycle risks, and generate optimized personalized sleep health risk assessment results. The optimized personalized sleep health risk assessment results include the future cycle risk change rate, the future cycle risk trend direction, and the predicted cycle risk index.

9. A personalized sleep health risk assessment method based on 5G edge computing according to claim 8, characterized in that, The specific steps of S5 are as follows: S501: Based on the personalized sleep health risk assessment results, extract the risk index sequence of continuous periods, calculate the difference between the risk index values ​​of adjacent periods, extract the risk change rate parameter according to the change direction and magnitude of the difference sequence, and generate a set of periodic risk change rates. S502: Call the set of periodic risk change rates, perform trend fitting operation on the multi-period risk change rates and corresponding risk indices, perform residual matching on the change rates and risk indices, determine the direction of risk change based on the sign of the slope of the fitted curve, and generate risk trend fitting results. S503: Based on the risk trend fitting results, predict and calculate the risk index for future periods, calculate the rate of change of the risk index for multiple prediction periods and weight and fuse it with the original risk index, update the periodic risk index, and generate an optimized personalized sleep health risk assessment result.

Images