An artificial intelligence-based health risk intelligent assessment method

Abstract

The application belongs to the technical field of health monitoring and data processing, and specifically discloses a health risk intelligent evaluation method based on artificial intelligence, which comprises the following steps: acquiring multiple physiological signals and performing time synchronization and adaptive wavelet packet noise reduction to generate synchronous physiological data, determining phase space reconstruction parameters by analyzing autocorrelation characteristics and adjacent point distribution changes, and mapping the physiological data into a physiological attractor manifold trajectory in a multidimensional phase space. In the initial learning stage, an individual reference manifold template is constructed, and normal state topological features are extracted. In the real-time evaluation stage, the current trajectory is extracted, the volume change amount and the local divergence rate are fused to calculate the deviation degree relative to the normal state. Finally, a health risk index is generated according to the deviation degree to trigger an early warning, and an environment adjustment instruction is output based on a reduced entropy compensation mapping model, so that accurate quantification of the health risk and environment closed-loop intervention are realized.

Description

Technical Field

[0001] This invention belongs to the field of health monitoring and data processing technology, and relates to an intelligent health risk assessment method based on artificial intelligence. Background Technology

[0002] Health risk assessment refers to the use of computer data processing technology to collect, analyze, and model human physiological signals to quantify the probability of an individual developing future health abnormalities. With the widespread adoption of wearable devices and IoT technology, it has become possible to continuously acquire physiological time-series data such as photoplethysmography (PPG) and acceleration. Artificial intelligence-based digital data processing methods, through feature extraction and pattern recognition of these massive amounts of continuous data, can reveal the dynamic evolutionary patterns hidden within complex physiological signals, providing individuals with continuous health status monitoring and early warning services.

[0003] Existing health risk assessment data processing solutions typically rely on extracting linear time-domain and frequency-domain features of physiological signals. The computer system calculates statistical indicators such as heart rate variability and pulse wave transit time within a fixed time window, then inputs these extracted features into a pre-trained machine learning classifier or threshold-based logic. When certain single-dimensional statistical features exceed a set static threshold, the system outputs a corresponding health risk warning and triggers simple environmental control device switching actions.

[0004] Existing data processing methods have inherent limitations when dealing with the complex nonlinear dynamics of the human body: fixed feature extraction methods are difficult to adapt to the dynamic changes in individual physiological states, and linear statistical indicators often ignore the topological evolution information of physiological signals in high-dimensional phase space, resulting in a lack of sensitivity to early, subtle nonlinear instability states. Furthermore, existing assessment results lack a refined mapping relationship based on physiological tolerance between them and environmental regulatory mechanisms; coarse intervention instructions can easily trigger abrupt changes in environmental parameters, failing to safely and smoothly guide the physiological system back to a healthy baseline.

[0005] To address the aforementioned issues, this invention employs multi-source physiological data synchronous denoising and phase space reconstruction techniques. It combines topological feature extraction and deviation calculation of high-dimensional manifold trajectories to generate a health risk index and introduces an entropy reduction compensation mapping model to output micro-incremental environmental regulation commands. This approach can accurately quantify the urgency of nonlinear instability in physiological systems and achieve safe and stable closed-loop environmental intervention. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art and to achieve the above objectives, the present invention proposes the following technical solution: a health risk intelligent assessment method based on artificial intelligence, comprising: S1, acquiring time series data of at least one physiological signal, performing time synchronization and adaptive noise reduction processing on the time series data, and generating synchronized physiological data.

[0007] S2. Based on synchronous physiological data, by analyzing the autocorrelation characteristics of the signal and the distribution changes of neighboring points in multidimensional space, phase space reconstruction parameters that match the current signal characteristics are determined. The phase space reconstruction parameters include delay time and embedding dimension.

[0008] S3. By utilizing the delay time and embedding dimension, delayed coordinate mapping is performed on the synchronous physiological data to convert the one-dimensional time series into a sequence of state points in a multi-dimensional phase space, forming a real-time physiological attractor manifold trajectory.

[0009] S4. In the initial learning phase, the real-time physiological attractor manifold trajectory is trained to construct an individual baseline manifold template, and the normal state topological features of the individual baseline manifold template are extracted by convex hull algorithm or clustering algorithm.

[0010] S5. In the real-time evaluation phase, extract the current physiological attractor manifold trajectory within the current sliding window and calculate the degree of deviation of the current physiological attractor manifold trajectory from the normal topological features.

[0011] S6. Generate a health risk index based on the degree of deviation. The health risk index is used to quantify the urgency of nonlinear instability in the physiological system.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention effectively filters out motion artifact interference and extracts the nonlinear dynamic basic features of the physiological system by performing time synchronization and adaptive wavelet packet denoising on multi-source physiological signals, and dynamically determining the phase space reconstruction parameters by combining autocorrelation characteristics and the distribution changes of adjacent points in multi-dimensional space. This data processing method avoids trajectory overlap or information loss caused by fixed parameters under complex physiological conditions, provides high-fidelity data support for subsequent high-dimensional phase space analysis, and improves the accuracy and robustness of digital data processing.

[0013] (2) This invention utilizes delayed coordinate mapping to transform a one-dimensional time series into physiological attractor manifold trajectories in a multi-dimensional phase space. It then constructs individual baseline manifold templates using convex hull or clustering algorithms to extract normal-state topological features such as hypervolume and distribution density. Finally, it integrates volume change and local divergence rate to generate the degree of deviation. This mechanism overcomes the limitations of traditional linear time-domain or frequency-domain analysis in capturing subtle nonlinear instabilities in physiological systems. It can objectively quantify the degree to which the current physiological state deviates from the healthy baseline from the perspective of high-dimensional geometry and topological structure breaks.

[0014] (3) After generating the health risk index, this invention introduces an entropy reduction compensation mapping model, which combines the current environmental parameters to generate micro-incremental step-by-step environmental adjustment instructions that have passed safety boundary checks, and continuously performs low-frequency monitoring and feedback during the execution process. This design realizes intelligent closed-loop control from risk calculation to physical environmental intervention. At the same time, through strict tolerance range verification and micro-step adjustment strategy, it prevents secondary stress shocks to the human body caused by drastic environmental changes, and ensures that the physiological system restores dynamic balance during a smooth transition. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the 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.

[0016] Figure 1 This is a schematic diagram of the implementation steps of the method of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 As shown, the present invention proposes an intelligent health risk assessment method based on artificial intelligence, comprising: S1, acquiring time series data of at least one physiological signal, performing time synchronization and adaptive noise reduction processing on the time series data, and generating synchronized physiological data.

[0019] In a preferred embodiment, time-series data of at least one physiological signal is acquired, and time-series data is subjected to time synchronization and adaptive noise reduction processing to generate synchronized physiological data. This includes: acquiring synchronized raw data containing photoplethysmography (PPG) signals and acceleration signals from a wearable device; resampling the PPG signals and acceleration signals to the same sampling frequency using a unified clock reference generated by a hardware timer to obtain a time-aligned signal; extracting motion features of the acceleration signal from the time-aligned signal, and performing adaptive wavelet packet noise reduction on the PPG signals in the time-aligned signal based on the motion features to remove motion-induced artifacts, thereby generating synchronized physiological data.

[0020] Specifically, the system acquires time-series data of physiological signals, including photoplethysmography (PPG) and acceleration signals. Physiological signals refer to electrical or optical parameters that reflect the internal functional state of the human body. Time-series data refers to a set of physiological parameter measurements arranged in chronological order. During the acquisition process, the system uses a photoelectric sensor built into the wearable device to collect raw optical signals reflecting changes in blood oxygen volume and a triaxial accelerometer to collect raw motion signals reflecting spatial displacement. Each raw signal is sampled by an analog-to-digital converter (ADC) and converted into digital time-series data.

[0021] The system performs time synchronization and adaptive noise reduction on time-series data to generate synchronized physiological data. Time synchronization refers to the process of eliminating time deviations between data from different sensors to achieve time axis alignment. Adaptive noise reduction refers to a method of automatically adjusting filtering parameters based on the dynamic characteristics of the signal itself to remove interference components. In specific implementation, the system acquires synchronized raw data containing photoplethysmography (PPG) and acceleration signals from a wearable device. A wearable device is an electronic device that fits against the human body to continuously collect human physiological indicators. A PPG signal is an electrical signal generated by detecting the periodic changes in blood volume with heartbeats using a photoelectric sensor built into the wearable device. An acceleration signal is an electrical signal reflecting the motion and vibration of the wearable device in space, acquired by a triaxial accelerometer built into the wearable device. Synchronized raw data refers to the unprocessed data set containing the above two signals collected within the same time period.

[0022] The device internally utilizes a unified clock reference generated by a hardware timer to resample the photoplethysmography (PPG) wave signal and acceleration signal to the same sampling frequency, resulting in a time-aligned signal. The hardware timer refers to the electronic component within the device used to generate fixed-frequency pulse signals. The unified clock reference refers to the reference signal that provides an absolute time reference for different sensors. The same sampling frequency means maintaining a consistent number of discrete samples extracted from the continuous signal within a set time span. During the resampling process, the system employs linear interpolation or cubic spline interpolation algorithms to complete the data points of the signal with the lower original sampling rate, or to decimate the signal with the higher sampling rate, ensuring that both signals have a data sequence with the same sampling interval on the time axis—this is the time-aligned signal.

[0023] The system extracts motion features from the acceleration signal from the time-aligned signal and performs adaptive wavelet packet denoising on the photoplethysmography (PPG) wave signal in the time-aligned signal based on these motion features to remove motion-induced artifacts, thereby generating synchronized physiological data. Motion features refer to numerical indicators reflecting the intensity and frequency distribution of human motion. Adaptive wavelet packet denoising is a data processing method that removes interference signals through multi-layer frequency band decomposition and threshold adjustment. Artifacts refer to non-physiological interference signals generated by the relative displacement between the photoelectric sensor and the skin caused by body movement. Synchronized physiological data refers to the data set that objectively reflects the human physiological state after time alignment and denoising. When extracting motion features, the system quantifies motion intensity by calculating the sliding window energy of the acceleration signal; the calculation formula is as follows:

[0024] In the formula, It represents the characteristics of motion and is a dimensionless quantity. This represents the total value of data points within the sliding window. It is a dimensionless quantity, and its definition is based on the product of the time span of the human movement cycle and the same sampling frequency. This represents the discrete sampling point numbers arranged in chronological order within the sliding window. ; Represents the time alignment signal of the first The acceleration signal sample values ​​at each sampling point are specifically the composite magnitudes of the X, Y, and Z axis acceleration components, with dimensions of [missing information]. ; Represents the reference acceleration, with dimensions of The setting is based on the measured value of gravitational acceleration of the human body in a resting state (usually taken as...). ).

[0025] In adaptive wavelet packet denoising, the system first uses a preset wavelet basis function to perform multi-level wavelet packet decomposition on the photoplethysmography (PPG) signal in the time-aligned signal, obtaining wavelet packet coefficients reflecting different frequency components of the signal. Multi-level wavelet packet decomposition refers to the operation of decomposing the signal into a series of frequency subspaces using orthogonal filter banks. Subsequently, the system dynamically calculates the denoising threshold based on motion characteristics, using the following formula: , in the formula Represents the denoising threshold, with dimensions of ; The modulating coefficient is a dimensionless quantity, and its setting is based on statistical analysis of multiple batches of noisy physiological signal samples. This represents the standard deviation of the photoplethysmography (PPG) signal within a specific frequency band, with dimensions of _____. The settings are based on the extraction of wavelet packet coefficients from the first-layer high-frequency detail nodes and the estimation of noise levels using the absolute median difference method. To prevent extreme value truncation caused by violent movement, when the calculated motion feature M exceeds the preset safety upper limit, it is forcibly limited to that safety upper limit to ensure that effective physiological signals are not excessively filtered out.

[0026] The system uses the calculated denoising threshold to perform soft thresholding on the wavelet packet coefficients of the photoplethysmography (PPG) signal. The calculation formula is as follows:

[0027] In the formula, The photoplethysmography signal represents the signal at the 1st... The denoised wavelet packet coefficients obtained after processing at each sampling point have dimensions of . ; The symbolic function represents a dimensionless quantity. The photoplethysmography (PPG) signal in the time-aligned signal is at the [missing information - likely a typo, should be 0]. The wavelet packet coefficients obtained after wavelet packet decomposition at each sampling point have dimensions of . ; This represents the function that maximizes the value. By combining with... Acceleration characteristics at the same index position By introducing threshold calculation, the noise reduction intensity is adaptively adjusted in real time according to the motion intensity.

[0028] The system uses the processed denoised wavelet packet coefficients to perform inverse wavelet packet transform reconstruction, restoring the denoised time-domain photoplethysmography (TPM) signal, and then combines it with a synchronized acceleration signal to generate synchronized physiological data. Inverse wavelet packet transform reconstruction refers to the computational process of mapping the transform domain coefficients back to the time domain to recover the signal waveform.

[0029] For example, assuming a user wears a wearable device while running, the system acquires time-series data of physiological signals, including photoplethysmography (PPG) and acceleration signals, through built-in photoelectric sensors and a triaxial accelerometer. The system performs time synchronization and adaptive noise reduction on the time-series data to generate synchronized physiological data. Specifically, the system acquires the synchronized raw data containing PPG and acceleration signals from the wearable device. Internally, using a unified clock reference generated by a hardware timer, the device employs a cubic spline interpolation algorithm to resample the PPG and acceleration signals to the same sampling frequency, set to 100Hz, resulting in a time-aligned signal. The system extracts the motion characteristics of the acceleration signal from the time-aligned signal. The total value of the data points within the sliding window is then set. 100, reference acceleration for Assume that within the current sliding window, the root mean square value of the triaxial synthesized magnitude sample of the acceleration signal at each sequence point in the time-aligned signal is... Substituting into the formula, the motion characteristics are calculated. The system performs adaptive wavelet packet denoising on photoplethysmography (PPG) signals in time-aligned signals based on motion characteristics. The system first uses a db4 wavelet basis for three-level decomposition to obtain the wavelet packet coefficients corresponding to each sampling point. And set the adjustment coefficient. The standard deviation of the photoplethysmography (PPG) signal in a specific frequency band is 0.5. The value is 0.2V. Substituting this value into the formula, the noise reduction threshold is calculated. The value is 0.4V. Assume the sequence points of the photoplethysmography (PPG) wave signal in the time-aligned signal are... wavelet packet coefficients The value is 0.6V. Substituting this value into the formula, we obtain the denoised wavelet packet coefficients after processing. The value is 0.2V. Finally, the system performs inverse wavelet packet transform on the processed coefficients to reconstruct the waveform, outputting the denoised time-domain pulse wave. This data directly verifies that by extracting motion features, dynamically adjusting the denoising threshold, and combining it with a complete wavelet packet decomposition and reconstruction process, artifact components can be filtered out, ensuring the reliability of synchronized physiological data.

[0030] S2. Based on synchronous physiological data, by analyzing the autocorrelation characteristics of the signal and the distribution changes of neighboring points in multidimensional space, phase space reconstruction parameters that match the current signal characteristics are determined. The phase space reconstruction parameters include delay time and embedding dimension.

[0031] In a preferred embodiment, based on synchronized physiological data, phase space reconstruction parameters matching the current signal characteristics are determined by analyzing the autocorrelation characteristics of the signal and the changes in the distribution of adjacent points in multidimensional space. This includes: calculating the autocorrelation function of the synchronized physiological data and determining the time interval corresponding to the first decrease of the autocorrelation coefficient to a preset proportion as the delay time; increasing the embedding dimension sequentially from a set initial dimension and calculating the proportion of point pairs in the phase space where adjacent state points separate after the dimension increase under different dimensions; when the proportion is lower than a preset threshold, determining the corresponding dimension as the minimum embedding dimension that can fully unfold the physiological dynamic trajectory, thereby obtaining the phase space reconstruction parameters.

[0032] Specifically, the system receives synchronized physiological data output from the preprocessing stage. Based on this data, it analyzes the autocorrelation characteristics of the signal and the changes in the distribution of neighboring points in the multidimensional space to determine phase space reconstruction parameters that match the current signal characteristics. The specific analysis method includes two sequential steps: First, by calculating the decay trend of the autocorrelation coefficient between the denoised photoplethysmography pulse wave signal and its time delay sequence, its autocorrelation characteristics are analyzed to extract the delay time that makes the phase space coordinate components sufficiently independent. Second, based on the determined delay time, the phase space is reconstructed, and the pseudo nearest neighbor (FNN) method is used to evaluate the proportion of adjacent state point pairs that separate during the spatial dimensionality upgrade process, analyze the changes in the distribution of neighboring points in the multidimensional space, and thus determine the minimum embedding dimension that can eliminate trajectory folding.

[0033] Synchronous physiological data refers to data sequences that, after time alignment and noise reduction, can objectively reflect the physiological state of the human body. In this step, the system specifically extracts the denoised photoplethysmography (PPG) signal from the synchronous physiological data as the analysis object for phase space reconstruction. Autocorrelation refers to the property of a signal's similarity to itself at different time delays. Changes in the distribution of adjacent points in multidimensional space refer to the change in distance between previously closely spaced data points as the spatial dimension increases after mapping a one-dimensional signal to a higher-dimensional space. Phase space reconstruction parameters refer to the set of control variables used to expand a one-dimensional time series into the state space of a higher-dimensional dynamic system. These parameters include delay time and embedding dimension. Delay time refers to the time interval between two adjacent coordinate components when constructing high-dimensional spatial coordinates. Embedding dimension refers to the minimum spatial dimension that can fully represent the dynamic characteristics of the system without trajectory overlap.

[0034] The system calculates the autocorrelation function of the denoised photoplethysmography (PPG) signal from synchronized physiological data. Autocorrelation function calculation refers to the process of mathematically evaluating the linear correlation between a signal sequence and its delayed sequence. The system obtains the autocorrelation coefficient using the following formula:

[0035] In the formula, Represents delay The autocorrelation coefficient for each sampling period is a dimensionless quantity. The autocorrelation coefficient is a numerical value that measures the similarity of a signal at different times. The first denoising photoplethysmography signal in the sequence represents the denoising photoplethysmography signal. Each sample value, with dimensions of . This represents the sample value in the sequence that is delayed by k sampling periods relative to the i-th sample value. The average value of the denoised photoplethysmography (PPG) signal sequence is represented by the dimensionless value. . It represents the total length of the denoised photoplethysmography (PPG) signal sequence. It is a dimensionless quantity and is defined based on measured data of human physiological cycle characteristics. The number of sampling points corresponding to the time interval is dimensionless. The system defines the time interval at which the autocorrelation coefficient first decreases to a preset proportion as the delay time. The preset proportion is a threshold used to determine whether the autocorrelation has decayed to a sufficiently small level to ensure the independence of the coordinate components; it is set based on statistical analysis of a large number of physiological signals in phase space. The time interval refers to the number of delay steps corresponding to when the autocorrelation coefficient reaches the preset proportion.

[0036] After determining the delay time, the system sets the initial embedding dimension to two dimensions (i.e., The system then progressively increases the embedding dimension from this initial dimension. As mentioned earlier, the system uses the pseudo nearest neighbor (FNN) method to determine the embedding dimension. The embedding dimension refers to the number of coordinate axes used when constructing the phase space. The system calculates the proportion of adjacent state points in the phase space that separate after the dimension is increased, under different dimensions. Adjacent state points are two data points in the phase space of a specific dimension whose Euclidean distance is less than a set threshold. The system calculates the distance between adjacent state points according to the following formula:

[0037] In the formula, This represents the embedding dimension as At that time, state point The Euclidean distance between it and its nearest neighbor state point has dimensions of ; Representing state points exist The starting index of the nearest neighbor state point in the signal sequence in dimensional space is a dimensionless quantity. The aforementioned delay time is a dimensionless quantity. This represents the dimension traversal index when calculating the Euclidean distance, with a value ranging from 0 to... Integers are dimensionless quantities. The sampled value at the target index position is obtained by shifting backward j times with the starting index i and the delay step number corresponding to τ as the step size in the signal sequence; The representative signal sequence is based on The sampled value at the target index position is obtained by shifting backward j times with the delay step number corresponding to τ as the starting index.

[0038] When the dimension is from Increase to At that time, the system uses the formula The formula calculates the proportion of point pairs that separate. Representing the proportion of point pairs that have separated, it is a dimensionless quantity. The proportion of point pairs that have separated refers to the ratio of the number of point pairs that were originally adjacent but are no longer adjacent due to the significant increase in distance after adding one dimension to the total number of adjacent point pairs. The number of point pairs that are no longer adjacent due to a significant increase in distance after the dimension is increased is a dimensionless quantity. The criterion for determining a significant increase in distance is: if the distance increases significantly after the dimension is increased... Distance from the original If the ratio is greater than the preset distance increase threshold, the point pair is determined to be a false nearest neighbor (i.e., a point pair that has separated); the distance increase threshold is set based on the statistical characteristics of the signal-to-noise ratio of physiological signals. Represents in dimension The total number of adjacent point pairs is a dimensionless quantity. When the proportion is lower than a preset threshold, the system determines the corresponding dimension as the minimum embedding dimension that can fully unfold the physiological dynamic trajectory. The preset threshold is the critical value for determining the proportion of false nearest neighbors that have been sufficiently eliminated, and it is set based on experimental measurements of multiple sets of nonlinear physiological system dynamics analysis. The minimum embedding dimension is the smallest dimensional value that makes the proportion of point pairs that separate below the preset threshold. The system thus obtains the phase space reconstruction parameters.

[0039] For example, the system receives synchronized physiological data, extracts the denoised photoplethysmography (PPG) signal, and determines phase space reconstruction parameters that match the current signal characteristics by analyzing the signal's autocorrelation characteristics and the distribution changes of neighboring points in multidimensional space. The system calculates the autocorrelation function of the denoised PPG signal. The total length of the synchronized physiological data sequence is set. The value is 1000. The system calculates the number of sampling points corresponding to different time intervals. Autocorrelation coefficient Set the preset ratio to 0.3. When As the autocorrelation coefficient gradually increases from 1 to 15, When the autocorrelation coefficient first drops to 0.28, which is below the preset ratio of 0.3, the system defines the time interval corresponding to the first drop of the autocorrelation coefficient to the preset ratio as the delay time. It is 15.

[0040] Subsequently, the system sets the initial embedding dimension to two dimensions (i.e., Starting from the initial dimension, the embedding dimension is increased sequentially, and the proportion of adjacent state points in the phase space that separate after the dimension increase is calculated at different dimensions. A preset threshold of 0.05 and a distance increase threshold of 10 are set. When the embedding dimension... When the dimension is 2, the system counts the total number of adjacent point pairs in dimension 2. The initial value was 500. After increasing the dimension to 3, the distance ratio caused by the increase in dimension exceeded the distance increase threshold, thus determining the number of point pairs that are no longer adjacent. The value is 200. Substituting this into the formula, we can calculate the proportion of point pairs that separated. The value was 0.4, higher than the preset threshold. The system further increased the embedding dimension to 3, and the total number of adjacent point pairs was calculated. The number of point pairs that are no longer adjacent due to their distance ratio exceeding a threshold after the dimension is increased from 300 to 4. The value is 12. Substituting this into the formula, we can calculate the proportion of point pairs that separated. The value is 0.04. At this point, the ratio is below the preset threshold of 0.05, and the system determines the corresponding dimension 3 as the minimum embedding dimension sufficient to fully unfold the physiological dynamic trajectory. The system thus obtains the phase space reconstruction parameters, including the delay time of 15 and the embedding dimension of 3. This data directly verifies that the phase space reconstruction parameters can be effectively determined through autocorrelation function calculation and adjacent state point separation ratio analysis, ensuring the accurate unfolding of the physiological dynamic trajectory.

[0041] S3. By utilizing the delay time and embedding dimension, delayed coordinate mapping is performed on the synchronous physiological data to convert the one-dimensional time series into a sequence of state points in a multi-dimensional phase space, forming a real-time physiological attractor manifold trajectory.

[0042] In a preferred embodiment, delayed coordinate mapping is performed on synchronized physiological data using delay time and embedding dimension to convert a one-dimensional time series into a state point sequence in a multi-dimensional phase space, forming a real-time physiological attractor manifold trajectory. This includes: delaying sampling of synchronized physiological data based on delay time and embedding dimension to construct multiple delayed coordinate vectors; arranging the multiple delayed coordinate vectors in chronological order to form a state point sequence in a high-dimensional phase space; and connecting the state point sequence in chronological order in the phase space to form a real-time physiological attractor manifold trajectory reflecting the current physiological dynamics.

[0043] Specifically, the system receives the delay time and embedding dimension determined in the preceding steps. Using the delay time and embedding dimension, the system performs delay coordinate mapping on the denoised photoplethysmography (PPG) signals in the synchronized physiological data. Delay coordinate mapping refers to a data transformation method that constructs a multidimensional spatial coordinate system from a univariate time series by introducing a time delay. The system converts the one-dimensional time series into a sequence of state points in a multidimensional phase space. In this step, the one-dimensional time series refers to the set of single denoised PPG parameter measurements arranged in chronological order. The multidimensional phase space refers to a mathematical space composed of multiple independent coordinate axes used to describe all possible states of the system. The sequence of state points refers to the set of coordinate points arranged in chronological order in the multidimensional space.

[0044] The system performs delayed sampling on the aforementioned one-dimensional time series in synchronized physiological data based on the delay time and embedding dimension. Delayed sampling refers to the operation of extracting discrete samples from a continuous data sequence at fixed time intervals. The system constructs multiple delayed coordinate vectors through delayed sampling. A delayed coordinate vector is a multi-dimensional spatial coordinate point composed of the current data and several historical delayed data points. The system constructs delayed coordinate vectors according to the formula:

[0045] In the formula, The representative sequence index is The delayed coordinate vector, with dimensions of ; The sequence index in the denoised photoplethysmography sequence is: The sampled values; It represents the delay time and is a dimensionless quantity. Represents the embedding dimension, which is the total dimension of the reconstructed phase space and is a dimensionless quantity; The state vector represents the state vector at the th The components in each dimension, that is, in the original signal sequence Starting at the index, offset backwards The sampled value at the position after a delay of τ steps.

[0046] The system arranges multiple delayed coordinate vectors in temporal order to form a sequence of state points in a high-dimensional phase space. A high-dimensional phase space refers to a mathematical space with dimensions greater than or equal to three used to represent the dynamic evolution of a system. The system follows the formula... The state point sequence is formed in the formula. Represents a sequence of state points, with dimensions of ; The total length of the state point sequence is a dimensionless quantity. It is calculated based on the difference between the total length of the one-dimensional time series, the delay time, and the embedding dimension. The specific calculation formula is as follows: ,in This represents the total length of the one-dimensional time series.

[0047] The system connects the sequence of state points in phase space in temporal order to form a real-time physiological attractor manifold trajectory reflecting the current physiological dynamics. The real-time physiological attractor manifold trajectory refers to a continuous curve with a specific geometric structure depicted by the evolution of system state points in phase space over time. The system forms the trajectory based on the following formula:

[0048] In the formula, Represents the real-time physiological attractor manifold trajectory, with dimensions of ; It represents the union operation of sets and is a dimensionless quantity. Represents a directed line segment connecting consecutive delayed coordinate vectors in phase space, with dimensions of .

[0049] For example, suppose the system receives synchronized physiological data and extracts the denoised photoplethysmography (PPG) signal as a one-dimensional time series, while simultaneously acquiring the delay time. The embedding dimension is 15. The system utilizes delay time and embedding dimension to perform delayed coordinate mapping on the signal. The system converts the one-dimensional time series into a sequence of state points in a multi-dimensional phase space. Based on the delay time and embedding dimension, the system performs delayed sampling on the signal, constructing multiple delayed coordinate vectors. Assume the sample value with sequence index 1 in the denoised opto-capacitive pulse wave sequence... The sample value is 0.5V with a sequence index of 16. The sample value is 0.6V with a sequence index of 31. The value is 0.4V. Substituting this into the formula, we obtain the delay coordinate vector for sequence index 1. for The dimension is V. The system continues to calculate the subsequent delay coordinate vector. The total length of the denoised photoplethysmography (PPG) sequence is set to 1000, and the total length of the state point sequence is calculated. The value is 970. The system arranges multiple delay coordinate vectors in time order to form a sequence of state points in a high-dimensional phase space. The system connects the state point sequence in phase space in time order, and assigns the sequence index 1 to the delay coordinate vector. The delay coordinate vector with sequence index 2 By connecting these connections and proceeding in sequence, a real-time physiological attractor manifold trajectory reflecting the current physiological dynamics is formed. This data directly verifies that delayed sampling and coordinate mapping can transform a one-dimensional time series into a trajectory in a multi-dimensional phase space, ensuring the complete reconstruction of the dynamic characteristics of the physiological system.

[0050] S4. In the initial learning phase, the real-time physiological attractor manifold trajectory is trained to construct an individual baseline manifold template, and the normal state topological features of the individual baseline manifold template are extracted by convex hull algorithm or clustering algorithm.

[0051] In a preferred embodiment, during the initial learning phase, the real-time physiological attractor manifold trajectory is trained to construct an individual baseline manifold template. The normal-state topological features of the individual baseline manifold template are then extracted using a convex hull algorithm or clustering algorithm. This includes: continuously acquiring real-time physiological attractor manifold trajectories for multiple time windows as training samples during the initial period when the user is in a resting state; processing the training samples using a convex hull algorithm or clustering algorithm to generate an individual baseline manifold template describing the outer envelope of the attractor in a normal state; and extracting the geometric and statistical features of the individual baseline manifold template, where the geometric and statistical features include the hypervolume and its distribution density on a preset cross-section, which together constitute the normal-state topological features.

[0052] Specifically, in the initial learning phase, the system trains on the real-time physiological attractor manifold trajectory generated in the preceding steps. The real-time physiological attractor manifold trajectory is specifically based on the denoised photoplethysmography pulse wave signal... The dynamic trajectory is reconstructed in phase space. The initial learning phase refers to the specific time interval used by the system to establish the user's personal physiological baseline before monitoring begins. During the initial period when the user is in a resting state, the system continuously acquires real-time physiological attractor manifold trajectories from multiple time windows as training samples. The initial resting state period refers to the starting time interval for data collection when the user is in a state of no significant physical activity and emotional stability. Multiple time windows refer to dividing continuous time into fixed-length time segments. Training samples refer to the set of real-time physiological attractor manifold trajectory data used to construct the baseline model.

[0053] The system uses convex hull or clustering algorithms to process training samples and generate individual baseline manifold templates describing the external envelope of the attractor in a normal state. Convex hull or clustering algorithms are mathematical computation methods used in multidimensional space to find boundaries that completely enclose given data points or to group similar data points. The external envelope of the attractor in a normal state refers to the boundary range that the dynamic trajectory of a physiological system in a healthy resting state can reach in phase space. The individual baseline manifold template is a mathematical model reflecting the current user's normal physiological dynamic boundary characteristics. Taking the convex hull algorithm as an example, the system constructs the individual baseline manifold template based on the formula:

[0054] In the formula, The representative baseline manifold template has the dimension V. It represents the total number of state points in the training samples. It is a dimensionless quantity and is determined based on the product of the length of multiple time windows and the sampling frequency. Represents the first in the training samples The term is a state point vector with the dimension V. The weight coefficients representing the corresponding state points are dimensionless quantities.

[0055] The system extracts the geometric and statistical features of individual baseline manifold templates. Geometric features refer to the mathematical properties describing the shape and size of the individual baseline manifold template in phase space. Statistical features refer to numerical indices reflecting the spatial distribution of state points within the individual baseline manifold template. Geometric and statistical features include hypervolume and its distribution density on a preset cross-section. Hypervolume refers to the size of the space occupied by a high-dimensional geometric object. The system calculates hypervolume using the following formula:

[0056] In the formula, Represents a hypervolume, with dimensions of . It represents the total number of simplexes that constitute the individual baseline manifold template and is a dimensionless quantity. It represents the embedding dimension determined by the preceding steps and is a dimensionless quantity. It represents matrix determinant operations and is a dimensionless quantity. Representing the The edge vector matrix of a simplex, with dimensions V, specifically representing the vectors pointing from one vertex of the simplex to the other vertices. vertices An m×m square matrix composed of vectors.

[0057] The distribution density on the preset cross section refers to the density of state points on a specified hyperplane in phase space. The preset cross section is a manually defined hyperplane that passes through the geometric center of the individual reference manifold template. The system calculates the distribution density on the preset cross section using the formula:

[0058] In the formula, Represents the distribution density on a preset cross section, with dimensions of . This represents the number of state points falling within the neighborhood of the preset cross section. It is a dimensionless quantity. The neighborhood refers to the spatial region where the Euclidean distance from the preset cross section is less than the preset thickness threshold. This represents the cross-sectional area of ​​the preset section within the individual reference manifold template, with dimensions of... .

[0059] The system extracts normal-state topological features from individual baseline manifold templates using convex hull or clustering algorithms. Convex hull or clustering algorithms are machine learning methods that automatically discover the inherent structure and patterns of data without manual labeling. Normal-state topological features refer to a set of features characterizing the dynamic structure of a healthy physiological system, consisting of the hypervolume and its distribution density on a predetermined cross-section.

[0060] For example, assuming the system enters the initial learning phase, during the initial period when the user is in a resting state, real-time physiological attractor manifold trajectories over multiple time windows are continuously acquired as training samples. The embedding dimension is then set. The system uses a convex hull algorithm or clustering algorithm to process training samples and generate individual baseline manifold templates describing the outer envelope of the attractor in the normal state. The system extracts the geometric and statistical features of the individual baseline manifold templates. The system calculates the hypervolume, assuming the total number of simplexes constituting the individual baseline manifold template. The edge vector matrix formed by the vertices of the simplex is 100. Given a 3×3 square matrix, after determinant operations and summation, the hypervolume is obtained by substituting into the formula. for The system calculates the distribution density on a preset cross-section, sets a preset thickness threshold of 0.01V, and assumes the number of state points falling within the neighborhood of the preset cross-section. The cross-sectional area of ​​the preset section within the individual reference manifold template is 450. for Substituting into the formula, the distribution density on the preset cross section is calculated. for The hypervolume and its distribution density on the preset cross-section together constitute the normal topological features. This data directly verifies that the geometric and statistical properties of individual baseline manifold templates can be effectively quantified through convex hull algorithms and feature extraction, ensuring the accurate construction of normal topological features.

[0061] S5. In the real-time evaluation phase, extract the current physiological attractor manifold trajectory within the current sliding window and calculate the degree of deviation of the current physiological attractor manifold trajectory from the normal topological features.

[0062] In a preferred embodiment, during the real-time evaluation phase, the current physiological attractor manifold trajectory within the current sliding window is extracted, and the deviation of the current physiological attractor manifold trajectory from the normal topological features is calculated. This includes: during real-time monitoring, using the sliding window to capture the latest physiological attractor manifold trajectory to form the current physiological attractor manifold trajectory; calculating the spatial volume occupied by the current physiological attractor manifold trajectory and comparing it with the hypervolume in the normal topological features to obtain the volume change; calculating the local divergence rate of the current physiological attractor manifold trajectory in phase space to obtain the trajectory divergence index; and fusing the volume change and the trajectory divergence index to generate the deviation characterizing the degree of topological structure disruption.

[0063] Specifically, during the real-time evaluation phase, the system extracts the current physiological attractor manifold trajectory within the current sliding window. The real-time evaluation phase refers to the operational phase where the system continuously monitors and analyzes the user's physiological state online after completing initial learning. The current sliding window refers to a data processing interval that moves at fixed steps across time-series data to extract data segments of fixed length. During real-time monitoring, the system uses the sliding window to extract the latest denoised photoplethysmography (PPG) signal and reconstructs its phase space using the delay time and embedding dimension determined in previous steps, thereby generating the current physiological attractor manifold trajectory. The sliding window refers to the time interval used to extract the latest data segment. The current physiological attractor manifold trajectory is a continuous curve formed by the evolution of the physiological system state points within the current sliding window time range in the aforementioned m-dimensional phase space.

[0064] The system calculates the spatial volume occupied by the current physiological attractor manifold trajectory. Spatial volume refers to the size of the geometric space enclosed by the current physiological attractor manifold trajectory in phase space. Specifically, the system uses the same convex hull algorithm as in the initial learning phase to construct the boundary envelope of the state point sequence within the current sliding window, and calculates the current spatial volume based on the constructed simplex. The system compares the calculated spatial volume with the hypervolume in the normal-state topological features to obtain the volume change. Normal-state topological features refer to the feature set extracted in the previous steps used to characterize the dynamic structure of the healthy physiological system. Hypervolume refers to the size of the space occupied by a high-dimensional geometry. The volume change is the relative difference between the spatial volume occupied by the current physiological attractor manifold trajectory and the hypervolume in the normal-state topological features. The system calculates the volume change using the formula:

[0065] In the formula, It represents the change in volume and is a dimensionless quantity. The dimensionless dimension represents the spatial volume occupied by the current physiological attractor manifold trajectory. . Represents the hypervolume in the normal topological features, with dimensions of .

[0066] The system calculates the local divergence rate of the current physiological attractor manifold trajectory in phase space. The local divergence rate refers to the exponential growth rate at which adjacent trajectories initially close in distance separate over time, used to quantify the chaotic characteristics of physiological system dynamics. The system calculates the local divergence rate using the formula:

[0067] In the formula, Represents the local divergence rate, with dimensions of . The number of evolution steps is a dimensionless quantity. It is determined based on one-tenth of the total number of data points contained in the sliding window length. When one-tenth of the total number of data points is less than 1, the value is set to 1 to ensure the validity of the number of evolution steps. Represents the sampling time interval, with dimensions of Its value is equal to the reciprocal of the aforementioned sampling frequency. During calculation, the system selects a reference state point within the current physiological attractor manifold trajectory and searches for the state point in phase space with the closest Euclidean distance to it as the adjacent trajectory point. This represents the initial Euclidean distance between the reference state point and its adjacent trajectory points, with dimensions V. This represents the two points mentioned above co-evolving in phase space for a specific number of steps. The Euclidean distance between the new positions is V. It represents the natural logarithm operation and is a dimensionless quantity.

[0068] The system acquires the trajectory divergence index. The trajectory divergence index is a comprehensive indicator that quantifies the local divergence rate of the current physiological attractor manifold trajectory in phase space. The system calculates the trajectory divergence index using the following formula:

[0069] In the formula, It represents the trajectory divergence index, which is a dimensionless quantity. Represents the reference divergence rate, with dimensions of The setting is based on the statistical average of the physiological trajectory divergence rate of healthy people in a resting state.

[0070] The system integrates volume change and trajectory divergence index to generate a deviation degree characterizing the topological structure disorder. Topological structure disorder refers to the disorder state of the physiological system's dynamic trajectory deviating from the geometry and topology of the healthy baseline manifold template. The degree of deviation is a comprehensive quantitative index obtained by integrating volume change and trajectory divergence index to measure the deviation of the current physiological state from the normal state. Although the normal state topological features also include the distribution density on the preset cross-section, in this step of real-time online evaluation, to improve the real-time performance of the system calculation and reduce computational power consumption, the macroscopic volume change and the trajectory divergence index reflecting dynamic stability are mainly selected to characterize the topological structure disorder. The distribution density on the preset cross-section is used as an offline auxiliary verification benchmark, which is used for secondary verification calculation at the system's bottom layer when the health risk index is in a critical warning state. The specific logic of the secondary review is as follows: When the health risk index is within a preset warning threshold range, the system triggers a secondary review judgment; comparing the current real-time distribution density with the statistical density of the individual baseline manifold template under the same cross-section, if the decrease in real-time distribution density exceeds a preset proportion threshold (e.g., 30%), the current topological defect is confirmed as a systemic trend shift, and the original warning index output is maintained; if it does not exceed the preset proportion threshold, the current shift is determined to be caused by transient random noise fluctuations, and smoothing filtering is performed on the current risk index to reduce the system's false alarm rate. The system generates the degree of deviation based on the formula:

[0071] In the formula, It represents the degree of deviation and is a dimensionless quantity. The weighting coefficient representing the change in volume is a dimensionless quantity, and its setting is based on expert assessment of the impact of physiological volume changes on health risks. The weighting coefficients representing the trajectory divergence index are dimensionless quantities. They are determined based on expert assessments of the impact of dynamic instability on health risks and satisfy the following conditions: The system calculates the degree of deviation of the current physiological attractor manifold trajectory from the normal topological features through the above steps.

[0072] For example, assuming the system enters the real-time evaluation phase, during real-time monitoring, the latest denoised photoplethysmography (PPG) signal is captured using a sliding window. This PPG signal is then reconstructed using an embedding dimension of 3 and a delay time of 15, forming the current physiological attractor manifold trajectory. The system then uses a convex hull algorithm to calculate the spatial volume occupied by the current physiological attractor manifold trajectory. for Hypervolumes in known normal topological features for Substituting into the formula, the volume change is calculated. It is 0.2.

[0073] The system calculates the local divergence rate of the current physiological attractor manifold trajectory in phase space. The number of evolution steps is set. The sampling time interval is 50. The value is 0.01s (corresponding to a 100Hz sampling frequency). The system selects a reference state point and tracks its evolution distance to its nearest neighbor. After logarithmic summation, assuming the accumulated result is 1.5, the local divergence rate is calculated using the formula. for The system acquires the trajectory divergence index. A reference divergence rate is set. for Substituting into the formula, the trajectory divergence index is calculated. It is 1.5.

[0074] The system integrates volume change and trajectory divergence index to generate a deviation characteristic of topological structure disruption. Weighting coefficients are set for the volume change. The weighting coefficient corresponding to the trajectory divergence index is 0.4. The value is 0.6. Substituting this into the formula, the degree of deviation is calculated. The value is 0.98. This data directly verifies that the degree of topological structure disruption can be accurately quantified by comparing spatial volume and calculating local divergence rate, ensuring that the degree of deviation can objectively reflect the deviation of the current physiological attractor manifold trajectory from the normal topological characteristics.

[0075] S6. Generate a health risk index based on the degree of deviation. The health risk index is used to quantify the urgency of nonlinear instability in the physiological system.

[0076] In a preferred embodiment, a health risk index is generated based on the degree of deviation. The health risk index is used to quantify the urgency of nonlinear instability in the physiological system, including: inputting the degree of deviation into a nonlinear mapping function and converting it into a normalized instability tightness value; combining the rate of change of the degree of deviation and weighting the instability tightness value to obtain the health risk index; and triggering an early warning output when the health risk index exceeds a preset early warning threshold.

[0077] In a further preferred embodiment, after generating the health risk index, the method further includes: acquiring current environmental parameters collected by environmental sensors; inputting the health risk index and the current environmental parameters into a pre-constructed entropy reduction compensation mapping model to generate an initial environmental adjustment command; performing a safety boundary check on the change range of the physical quantity corresponding to the initial environmental adjustment command; if the change range exceeds the human body's tolerance range, stopping the command output; otherwise, outputting a verified environmental adjustment command; and decomposing the verified environmental adjustment command into an execution message containing a micro-increment step size and a time interval, and sending it to the corresponding environmental adjustment actuator.

[0078] In a further preferred embodiment, after sending to the corresponding environmental regulation actuator, the method further includes: after the environmental regulation actuator performs each step action, re-acquiring synchronized physiological data and calculating a new health risk index; when the newly calculated health risk index returns to the preset safe range, stopping the generation of subsequent environmental regulation instructions and putting the system into a low-frequency monitoring mode.

[0079] Specifically, the system generates a health risk index based on the deviation calculated in previous steps. The health risk index quantifies the urgency of nonlinear instability in the physiological system. Nonlinear instability refers to a dangerous state where the dynamic balance within the human physiological system is disrupted and cannot be self-recovered. The system inputs the deviation degree into a nonlinear mapping function, transforming it into a normalized instability severity value. The nonlinear mapping function is a mathematical model that maps input data to a specific interval through nonlinear mathematical relationships. The instability severity value, after nonlinear mapping, is a numerical value between 0 and 1 representing the risk of instability. The system calculates the instability severity value using the following formula:

[0080] In the formula, It represents the instability tightness value and is a dimensionless quantity. It represents the natural constant and is a dimensionless quantity. This represents the mapping steepness coefficient, which is a dimensionless quantity and is determined based on fitting data from physiological instability samples. It represents the degree of deviation and is a dimensionless quantity. This represents a deviation from the baseline threshold and is a dimensionless quantity. It is set based on the statistical upper limit of the deviation degree of healthy people.

[0081] The system combines the rate of change of deviation with a weighted adjustment of the instability tightness value to obtain a health risk index. The rate of change of deviation refers to the increment of deviation per unit time. The system calculates the rate of change of deviation using the following formula:

[0082] In the formula, The rate of change representing the degree of deviation, with dimensions of . It represents the degree of deviation from the previous evaluation period and is a dimensionless quantity. Represents the time interval of the evaluation cycle, with the dimension of .

[0083] The system calculates the health risk index using the following formula:

[0084] In the formula, It represents a health risk index and is a dimensionless quantity. It represents the compactness weighting coefficient and is a dimensionless quantity. The representative rate weighting coefficient is a dimensionless quantity, and its setting is based on expert experience evaluation. Represents the time normalization constant, with dimensions of The basis for this is to eliminate the dimension of rate in order to achieve numerical addition.

[0085] When the health risk index exceeds a preset warning threshold, the system triggers a warning output. The preset warning threshold refers to the critical value of the health risk index that triggers the system alarm, and it is set based on statistical analysis of a large amount of clinical critical illness warning data. The warning output refers to the danger alert signal sent by the system to the user or monitoring equipment.

[0086] After generating the health risk index, the system acquires the current environmental parameters collected by environmental sensors. Environmental sensors are electronic components used to detect the physical state of the surrounding environment. Current environmental parameters refer to the real-time values ​​of temperature or humidity collected by the environmental sensors. The following formulas and explanations use temperature regulation as an example. The system inputs the health risk index and current environmental parameters into a pre-constructed entropy reduction compensation mapping model to generate initial environmental regulation instructions. The entropy reduction compensation mapping model is a mathematical control model that adjusts environmental parameters to reduce the disorder of the human physiological system and restore stability. Its core logic is to simultaneously consider feedforward risk compensation dominated by physiological indicators and feedback error correction dominated by environmental indicators. Specifically, the entropy reduction compensation mapping model internally stores a correlation matrix between risk intervals and physical regulation weights. Based on the level interval to which the health risk index belongs (e.g., low risk, medium risk, high risk), the model automatically matches the corresponding combination of environmental factor adjustments, thereby converting the dimensionless risk index into specific physical quantity regulation instructions. For example, when the risk index indicates increased physiological stress, the mapping model outputs a set of correlated physical instructions including reducing the ambient temperature step size, increasing the fresh air volume weight, and adjusting the light color temperature. The initial environmental adjustment command refers to the control signal derived from the preliminary calculations of the model to change the environmental state. The system calculates the magnitude of the physical quantity change corresponding to the initial environmental adjustment command based on the formula:

[0087] In the formula, This represents the magnitude of the physical quantity change corresponding to the initial environmental adjustment command, with the dimension being ℃ (or %RH if it is humidity adjustment, the same below). Represents the risk compensation coefficient, with the dimension of ℃. It is set based on the sensitivity test of human physiology to environmental changes and is used to directly output the environmental counter-regulation amount according to the health risk index. It represents the environmental feedback coefficient and is a dimensionless quantity. This represents the current environmental parameters, measured in °C. This represents the optimal environmental parameter, measured in °C, and is based on human comfort standards. The latter part of the formula... It is used to correct errors caused by the current environment deviating from the optimal comfort zone.

[0088] The system performs a safety boundary check on the magnitude of changes in physical quantities corresponding to the initial environmental adjustment command. The safety boundary check verifies whether the magnitude of the physical quantity change conforms to the limits of human physiological tolerance. If the magnitude exceeds the human tolerance range, the system halts command output and triggers a manual intervention alarm to prevent secondary risks arising from automated adjustment. The human tolerance range refers to the range of environmental parameter changes that the human body can safely adapt to. Otherwise, the system outputs a verified environmental adjustment command. A verified environmental adjustment command is a control signal that is allowed to be executed after passing the safety boundary check. The system judges the absolute value. Is it less than or equal to the maximum range of variation that the human body can tolerate? If satisfied, then The magnitude of the physical quantity change corresponding to the verified environmental control command The system decomposes the verified environmental adjustment instructions into execution messages containing micro-increment step sizes and time intervals. The micro-increment step size refers to the tiny, single execution amount into which the total adjustment is divided. The time interval refers to the waiting time between two adjacent micro-increment step executions. An execution message is a data packet containing the specific parameters of the control instruction. Furthermore, to avoid hardware oscillations caused by control signals, the total number of execution steps is limited. With time interval The settings need to be adapted to the minimum physical response cycle of the actuator in the corresponding environment. It should be noted that the time interval in the execution message... The numerical setting is usually greater than the evaluation cycle time interval. This design aims to match the physical response inertia of hardware devices, ensuring the smoothness of the environmental incremental intervention process and effectively preventing oscillations in controlled environmental parameters caused by excessively frequent adjustment actions. The system calculates the incremental step size using the following formula:

[0089] In the formula, It represents the incremental step size, with dimensions in °C. This represents the total number of execution steps and is a dimensionless quantity. Its setting is based on the control accuracy of the environmental control equipment. The system calculates the time interval using the formula:

[0090] In the formula, Represents a time interval, with dimensions of . Represents the total settling time, with dimensions of The system will send the execution message to the corresponding environmental control actuator. An environmental control actuator is a hardware device that receives instructions and changes the environmental state.

[0091] After the data is sent to the corresponding environmental regulation actuator, the system reacquires synchronized physiological data and calculates a new health risk index after each step of the actuator's operation. Each step refers to the environmental regulation actuator completing a single micro-increment step. During this process, the system acquires new sliding window data over time and recalculates the health risk index by sequentially going through the feature extraction and phase space reconstruction processes described in steps S1 to S5. This new health risk index is the final health risk index. When the newly calculated health risk index returns to the preset safety range, the system stops generating subsequent environmental regulation commands and enters a low-frequency monitoring mode. The preset safety range refers to the index range representing a stable and healthy physiological system, set based on daily physiological monitoring data of healthy individuals. Subsequent environmental regulation commands refer to the remaining environmental regulation commands that have not yet been executed. The low-frequency monitoring mode refers to an operating state that reduces the frequency of data acquisition and processing to save system energy.

[0092] For example, suppose the system generates a health risk index based on the degree of deviation. The degree of deviation is known. The value is 0.98. Set the mapping steepness coefficient. The value is 10, which deviates from the baseline threshold. The value is 0.8. Substituting this into the formula, the instability tightness value is calculated. The value is 0.86. The deviation from the previous evaluation period is known. The value is 0.93, and the evaluation period interval is [missing information]. The value is 5 seconds. Substituting this value into the formula, we can calculate the rate of change of the deviation. for Set the tightness weighting coefficient. The rate weighting coefficient is 0.7. The time normalization constant is 0.3. The value is 1 second. Substituting this value into the formula, the health risk index is calculated. The value is 0.605. The preset warning threshold is set to 0.8, and the current health risk index does not exceed this threshold. The system acquires current environmental parameters collected by environmental sensors. for Set a risk compensation coefficient. for Environmental feedback coefficient for Optimal environmental parameters for Substituting the values ​​into the formula, we can calculate the magnitude of the physical quantity change corresponding to the initial environmental adjustment command. for Set the maximum range of variation within the human body's tolerance. for Due to absolute value Less than The system outputs verified environmental adjustment commands, and the corresponding changes in physical quantities are shown. for Set the total number of execution steps. The total adjustment time is 10. The value is 300s. Substituting this into the formula, the incremental step size is calculated. for Time interval The time limit is 30 seconds. The system sends an execution message containing the above parameters to the corresponding environmental regulation actuator. After each step action is performed by the environmental regulation actuator, the system reacquires synchronized physiological data based on the latest sliding window and performs phase space reconstruction and deviation calculation to calculate a new health risk index. Assuming that after 5 steps, the newly calculated health risk index drops to 0.3, falling within the preset safe range of 0 to 0.4, the system stops generating subsequent environmental regulation commands and enters a low-frequency monitoring mode. This data directly verifies that the health risk index calculation and entropy reduction compensation mapping model can effectively regulate environmental parameters and promote the physiological system to return to stability, realizing closed-loop control from physiological monitoring to environmental intervention.

[0093] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A health risk intelligent assessment method based on artificial intelligence, characterized in that, include: S1. Acquire time-series data of at least one physiological signal, perform time synchronization and adaptive noise reduction on the time-series data, and generate synchronized physiological data; S2. Based on synchronous physiological data, by analyzing the autocorrelation characteristics of the signal and the distribution changes of neighboring points in multidimensional space, phase space reconstruction parameters that match the current signal characteristics are determined. The phase space reconstruction parameters include delay time and embedding dimension. S3. By utilizing the delay time and embedding dimension, delayed coordinate mapping is performed on the synchronous physiological data to convert the one-dimensional time series into a sequence of state points in a multi-dimensional phase space, forming a real-time physiological attractor manifold trajectory. S4. In the initial learning phase, the real-time physiological attractor manifold trajectory is trained to construct an individual baseline manifold template, and the normal state topological features of the individual baseline manifold template are extracted by convex hull algorithm or clustering algorithm. S5. In the real-time evaluation phase, extract the current physiological attractor manifold trajectory within the current sliding window and calculate the degree of deviation of the current physiological attractor manifold trajectory from the normal topological features. S6. Generate a health risk index based on the degree of deviation. The health risk index is used to quantify the urgency of nonlinear instability in the physiological system.

2. The intelligent health risk assessment method based on artificial intelligence according to claim 1, characterized in that, Acquire time-series data of at least one physiological signal, perform time synchronization and adaptive noise reduction processing on the time-series data, and generate synchronized physiological data, including: Acquire synchronous raw data containing photoplethysmography (PPG) signals and acceleration signals from wearable devices; By using a unified clock reference generated by a hardware timer, the photoplethysmography (PPG) signal and acceleration signal are resampled to the same sampling frequency to obtain a time-aligned signal. Motion features of acceleration signals are extracted from time-aligned signals, and adaptive wavelet packet denoising is performed on photoplethysmography pulse wave signals in time-aligned signals based on motion features to remove motion-induced artifacts, thereby generating synchronized physiological data.

3. The intelligent health risk assessment method based on artificial intelligence according to claim 1, characterized in that, Based on synchronized physiological data, by analyzing the autocorrelation characteristics of the signal and the changes in the distribution of neighboring points in multidimensional space, phase space reconstruction parameters that match the characteristics of the current signal are determined, including: The autocorrelation function of the synchronous physiological data is calculated, and the time interval corresponding to the first decrease of the autocorrelation coefficient to a preset ratio is determined as the delay time. Starting from the initial dimension, the embedding dimension is increased dimension by dimension, and the proportion of adjacent state points in the phase space that separate after the dimension is increased is calculated under different dimensions. When the proportion is lower than the preset threshold, the corresponding dimension is determined as the minimum embedding dimension that can fully unfold the physiological dynamic trajectory, thereby obtaining the phase space reconstruction parameters.

4. The intelligent health risk assessment method based on artificial intelligence according to claim 1, characterized in that, By utilizing time delay and embedding dimension, delayed coordinate mapping is performed on synchronized physiological data, transforming a one-dimensional time series into a sequence of state points in a multi-dimensional phase space, forming a real-time physiological attractor manifold trajectory, including: Based on the delay time and embedding dimension, synchronous physiological data are sampled at a delay time to construct multiple delayed coordinate vectors; Arrange multiple delayed coordinate vectors in time order to form a sequence of state points in a high-dimensional phase space; By connecting the state point sequence in phase space in chronological order, a real-time physiological attractor manifold trajectory reflecting the current physiological dynamics is formed.

5. The intelligent health risk assessment method based on artificial intelligence according to claim 1, characterized in that, In the initial learning phase, the real-time physiological attractor manifold trajectory is trained to construct an individual baseline manifold template. Normal-state topological features of the individual baseline manifold template are then extracted using convex hull or clustering algorithms, including: During the initial period when the user is in a resting state, real-time physiological attractor manifold trajectories of multiple time windows are continuously acquired as training samples. The training samples are processed using convex hull or clustering algorithms to generate individual baseline manifold templates that describe the outer envelope of the attractor in the normal state. Geometric and statistical features of individual baseline manifold templates are extracted. These features include hypervolume and distribution density on a preset cross section, which together constitute normal topological features.

6. The intelligent health risk assessment method based on artificial intelligence according to claim 5, characterized in that, During the real-time evaluation phase, the current physiological attractor manifold trajectory within the current sliding window is extracted, and the deviation of the current physiological attractor manifold trajectory from the normal topological features is calculated. This includes: during real-time monitoring, the latest physiological attractor manifold trajectory is captured using the sliding window to form the current physiological attractor manifold trajectory. Calculate the spatial volume occupied by the current physiological attractor manifold trajectory, compare it with the hypervolume in the normal topological features, and obtain the volume change. Calculate the local divergence rate of the current physiological attractor manifold trajectory in phase space and obtain the trajectory divergence index; By integrating volume change and trajectory divergence index, a deviation characteristic of topological structure breaking is generated.

7. The intelligent health risk assessment method based on artificial intelligence according to claim 1, characterized in that, A health risk index is generated based on the degree of deviation. This health risk index is used to quantify the urgency of nonlinear instability in the physiological system, including: The degree of deviation is input into a nonlinear mapping function and transformed into a normalized instability tightness value; By combining the rate of change of the degree of deviation, the instability tightness value is weighted and adjusted to obtain the health risk index; When the health risk index exceeds the preset warning threshold, a warning is triggered.

8. The intelligent health risk assessment method based on artificial intelligence according to claim 7, characterized in that, After generating the health risk index, the following is also included: Acquire current environmental parameters collected by environmental sensors; The health risk index and current environmental parameters are input into a pre-built entropy reduction compensation mapping model to generate initial environmental adjustment instructions. The change range of the physical quantity corresponding to the initial environmental adjustment command is checked for safety boundaries. If the change range exceeds the human body's tolerance range, the command output is stopped; otherwise, the verified environmental adjustment command is output. The verified environmental control command is decomposed into an execution message containing the incremental step size and time interval, and sent to the corresponding environmental control actuator.

9. The intelligent health risk assessment method based on artificial intelligence according to claim 8, characterized in that, After being sent to the corresponding environmental conditioning actuator, it also includes: After the environmental regulation actuator performs each step, synchronized physiological data is reacquired and a new health risk index is calculated. When the newly calculated health risk index returns to the preset safe range, the generation of subsequent environmental adjustment commands will stop, and the system will enter a low-frequency monitoring mode.

Images