Biological feedback equipment data processing method based on artificial intelligence

Abstract

The invention discloses a biological feedback equipment data processing method based on artificial intelligence, relates to the technical field of biological feedback data processing, and realizes accurate decoupling of respiratory rhythm and autonomic nerve regulation signals in a nonlinear feature space through a submerged space orthogonal decomposition mechanism. The problem of HRV low-frequency-band information loss caused by traditional frequency domain filtering is avoided; the respiration component estimation based on the variational auto-encoder does not depend on an external sensor, the hardware cost and the user burden are reduced, and meanwhile, the stability of long-term analysis is guaranteed by self-adaption to individual respiration mode difference and frequency drift through a dynamic orthogonal constraint and online fine adjustment mechanism.

Description

Technical Field

[0001] The present invention relates to the technical field of biofeedback data processing, and in particular to a biofeedback device data processing method based on artificial intelligence. Background Art

[0002] Heart rate variability (HRV), as a core indicator for assessing the function of the autonomic nervous system, is widely used in biofeedback devices in areas such as stress monitoring, mood regulation, and cardiovascular disease early warning. Traditional HRV analysis relies on frequency domain analysis of RR interval signals, extracting low-frequency (LF) (0.04-0.15Hz) and high-frequency (HF) (0.15-0.4Hz) components through fast Fourier transform (FFT) or wavelet transform to quantify the balance between the sympathetic and parasympathetic nerves.

[0003] However, the complexity of human physiological signals makes HRV analysis susceptible to interference from respiratory rhythm. Respiratory movement, through mechanical stretch reflexes and central nervous system coupling, triggers periodic fluctuations in the RR interval. These fluctuations (0.1-0.3 Hz) overlap significantly with the LF band of HRV. This spectral confusion makes it difficult for traditional methods to distinguish between autonomic regulatory signals and respiratory rhythm coupling noise, reducing the clinical interpretability of HRV features. To suppress respiratory interference, some early biofeedback devices employed two strategies: frequency-domain bandpass filtering, which uses a fixed cutoff frequency to filter out low-frequency respiratory components. However, this also loses critical information in the LF band of HRV, distorting autonomic function assessment. The other is multi-sensor fusion, which uses independent respiratory sensors to synchronously acquire respiratory signals and uses linear regression or blind source separation algorithms to remove respiratory-related components from the heart rate signal. However, due to individual differences in respiratory frequency, it is difficult to pre-set universal filtering parameters for the former, while the latter requires increased hardware cost and wearable burden for the user. Moreover, linear models cannot account for the nonlinear dynamic coupling relationship between respiration and HRV.

[0004] However, the impact of breathing on HRV is not a simple spectral superposition or linear modulation, but a non-stationary dynamic process formed by a neuro-mechanical feedback loop. For example, in the rapid shallow breathing state of anxiety patients, the respiratory rate may briefly invade the LF frequency band. At this time, traditional FFT analysis will mistakenly attribute the respiratory rhythm to sympathetic hyperactivity, misleading treatment decisions; although some studies have attempted to introduce adaptive filtering or time-frequency analysis (such as Wigner-Ville distribution), its high computational complexity and sensitivity to noise make it difficult to meet the real-time requirements of biofeedback equipment; therefore, it is urgent to construct an accurate decoupling model of the respiratory-HRV coupling relationship from the algorithm level, and realize sensorless and low-complexity respiratory interference suppression while retaining the physiological significance of HRV, providing a technical basis for high-precision autonomic nervous function assessment. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides an artificial intelligence-based biofeedback device data processing method to solve the problems that traditional HRV analysis is interfered by respiratory rhythm coupling, frequency domain filtering causes signal distortion, multi-sensor solutions increase the hardware burden, and linear models cannot analyze dynamic nonlinear coupling relationships.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] An embodiment of the present invention provides a biofeedback device data processing method based on artificial intelligence, which includes:

[0009] Step S1, obtaining the user's electrocardiogram (ECG) signal and respiratory signal, wherein the respiratory signal is generated by a chest and abdomen motion sensor or estimated based on periodic fluctuation characteristics of the ECG signal;

[0010] Step S2, converting the ECG signal into a RR interval sequence and inputting it into a variational autoencoder for latent space decomposition to obtain a latent space representation including a respiratory rhythm component and a neural regulation component;

[0011] Step S3, performing orthogonal processing on the latent space representation so that the respiratory rhythm component and the neural regulation component are orthogonal in the latent space;

[0012] Step S4, reconstructing the RR interval sequence without non-respiratory coupling based on the orthogonalized latent space representation;

[0013] Step S5: Calculate frequency domain characteristic parameters based on the reconstructed RR interval sequence and output them to the visualization interface of the biofeedback device.

[0014] As a preferred embodiment of the data processing method for a biofeedback device based on artificial intelligence of the present invention, the step of estimating and generating the respiratory signal further comprises:

[0015] Adaptive sliding window Fourier transform was performed on the RR interval sequence to extract the amplitude change characteristics in the 0.1-0.3 Hz frequency band;

[0016] A simulated respiratory waveform is generated based on the amplitude variation characteristic as an input substitute for the respiratory signal.

[0017] As a preferred embodiment of the data processing method of the biofeedback device based on artificial intelligence described in the present invention, the latent space decomposition step of the variational autoencoder includes:

[0018] Constructing the latent variable subspace of respiratory rhythm component and the latent variable subspace of neural regulation component;

[0019] The distribution of the respiratory rhythm component subspace is constrained to match the time correlation characteristics of the respiratory signal through adversarial training.

[0020] As a preferred embodiment of the data processing method of the biofeedback device based on artificial intelligence of the present invention, in step S2, the RR interval sequence and latent space decomposition are generated in the following manner:

[0021] The Pan-Tompkins algorithm is used to detect the R wave peak on the original ECG signal x(t) to obtain the peak time sequence And calculate the RR interval: in, Indicates the (n+1)th R peak moment, Indicates the nth R peak moment, RR n represents the nth RR interval;

[0022] {RR n}After normalization, construct column vector RR=[RR1,RR2,…,RR N ] T , where RR represents the input RR interval sequence column vector and N represents the sequence length; it is fed into the variational autoencoder VAE encoder to obtain the latent space representation: Where z∈R d represents the overall vector of the latent space, represents the encoder mapping, For its parameters;

[0023] Divide z into respiratory rhythm component subspace and neural regulation component subspace according to dimension:

[0024]

[0025] Among them, z r Represents the respiratory rhythm subspace vector with dimension d r , a n Represents the neural modulation subspace vector with dimension d n , d represents the total dimension of the latent space, and R here represents the set of real numbers;

[0026] Introduce adversarial training and define adversarial loss as:

[0027]

[0028] Among them, L adv represents the adversarial training loss, Represents the real breathing feature sample x r expectations, Represents the encoder output breathing vector z r Expectations, Aφ (·) represents the adversarial discriminator mapping, φ is the discriminator parameter, x r Represents the representation of the real breathing signal in the feature domain, z r represents the breathing subspace vector generated by the encoder;

[0029] The encoder and discriminator pass:

[0030] The confrontation process makes z r The distribution of x is gradually aligned r distribution.

[0031] As a preferred embodiment of the data processing method of the biofeedback device based on artificial intelligence described in the present invention, the orthogonalization processing step includes:

[0032] Calculate the principal component direction of the respiratory rhythm component in the latent space;

[0033] An orthogonal projection constraint is imposed on the neuromodulation component so that the angle between it and the principal component direction is not less than 80 degrees.

[0034] As a preferred solution of the artificial intelligence-based biofeedback device data processing method of the present invention, in step S3, the step of orthogonalizing the latent space is as follows:

[0035] Calculate the covariance matrix of the respiratory rhythm subspace vector set, the formula is: C r =Cov(z r ),in, Represents the breathing subspace vector z r The covariance matrix of , Cov(·) is the covariance operator;

[0036] On this basis, by solving the principal component analysis, the main direction vector u of the breathing subspace is obtained:

[0037]

[0038] in, represents the principal component direction vector, |v| represents the Euclidean norm of vector v;

[0039] Perform orthogonal projection on the neuromodulation component vector to obtain the orthogonalized vector:

[0040] z' n =z n -(u T z n )u, where represents the orthogonalized neural modulation vector, u T z n represents the vector inner product;

[0041] Define the orthogonality penalty term:

[0042] L orth =max(0,|u T z n |-|z n |cos(80°)),

[0043] Among them, L orth represents the orthogonality penalty loss, max(·) is the maximum value operator, |z n | represents the Euclidean norm of the neuromodulation vector, cos(80°) represents the cosine value of 80 degrees;

[0044] This term is incorporated into the overall training loss to suppress overlap between subspaces.

[0045] As a preferred solution of the artificial intelligence-based biofeedback device data processing method of the present invention, in step S3, an orthogonal projection constraint is applied to the neuromodulation component, specifically:

[0046] The angle between vectors is defined as:

[0047]

[0048] in, represents the principal component direction vector of the breathing subspace, represents the neuromodulatory component vector, |·| represents the Euclidean norm, arccos(·) is the arccosine function, and θ is the angle between vectors;

[0049] The geometric constraint is written as θ≥80°, which is equivalent to:

[0050] Where cos(80°) represents the cosine value of 80°;

[0051] In actual training, a soft constraint penalty term is used:

[0052]

[0053] Among them, L orth is the orthogonality penalty loss, max(·) takes the maximum value operator;

[0054] The threshold setting logic includes: in HRV analysis, 80° corresponds to a cosine value of approximately 0.1736, which ensures a high degree of subspace separation without excessively weakening the neural features; a cosine threshold that is too small will lead to unstable training, while a cosine threshold that is too large will result in insufficient separation. 80° is the balance point between performance and robustness in experience; in different individuals or application scenarios, 80° can be used as the initial value and fine-tuned according to the performance of the validation set.

[0055] As a preferred solution of the artificial intelligence-based biofeedback device data processing method of the present invention, the step of calculating the frequency domain characteristic parameters includes:

[0056] The reconstructed RR interval sequence was corrected in the frequency domain, specifically by dynamically suppressing the power density value in the region overlapping with the respiratory estimation frequency band within the LF frequency band of 0.04-0.15 Hz.

[0057] As a preferred embodiment of the data processing method of the biofeedback device based on artificial intelligence described in the present invention, it further includes a dynamic adjustment step:

[0058] Monitoring the frequency drift of the respiratory signal, and triggering online fine-tuning of the parameters of the variational autoencoder when the frequency drift exceeds a preset threshold;

[0059] The online fine-tuning adopts an unsupervised loss function and updates the model weights only based on the orthogonality constraint of the latent space of the current input signal.

[0060] As a preferred solution of the artificial intelligence-based biofeedback device data processing method described in the present invention, the method is deployed on an edge computing device and specifically includes:

[0061] Compiling the variational autoencoder and the orthogonalization processing step into a lightweight inference engine; the lightweight inference engine performs model compression through structured pruning and 8-bit integer quantization;

[0062] The RR interval sequence is streamed using a memory mapping technique that performs zero-copy data transfer based on a DMA controller of a target edge device.

[0063] The beneficial effects of the present invention are as follows: the present invention realizes the precise decoupling of respiratory rhythm and autonomic nervous regulation signal in the nonlinear feature space through the latent space orthogonal decomposition mechanism, thereby avoiding the problem of HRV low-frequency band information loss caused by traditional frequency domain filtering; respiratory component estimation based on variational autoencoder does not need to rely on external sensors, reducing hardware costs and user burden, and at the same time, through dynamic orthogonal constraints and online fine-tuning mechanisms, it adapts to individual respiratory pattern differences and frequency drift, ensuring the stability of long-term analysis; lightweight compilation and streaming processing technology in edge deployment enables complex models to run in real time on low-power devices, overcoming the defects of computational redundancy of traditional adaptive algorithms; while retaining the physiological significance of HRV, the robustness of respiratory interference suppression and system applicability are significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0065] Figure 1 Schematic diagram of the process of biofeedback data processing in Example 1. DETAILED DESCRIPTION

[0066] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0067] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0068] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0069] Example 1, reference Figure 1 , this embodiment provides a biofeedback data processing method based on artificial intelligence, comprising:

[0070] Step S1, obtaining the user's electrocardiogram (ECG) signal and respiratory signal, wherein the respiratory signal is generated by a chest and abdomen motion sensor or estimated based on periodic fluctuation characteristics of the ECG signal;

[0071] The step of generating an estimate of the breathing signal further comprises:

[0072] Adaptive sliding window Fourier transform was performed on the RR interval sequence to extract the amplitude change characteristics in the 0.1-0.3 Hz frequency band;

[0073] Generate a simulated respiratory waveform based on the amplitude change characteristics as an alternative input of the respiratory signal;

[0074] Step S2: converting the ECG signal into a RR interval sequence and inputting it into a variational autoencoder for latent space decomposition to obtain a latent space representation containing respiratory rhythm components and neural regulation components;

[0075] The latent space decomposition steps of the variational autoencoder include:

[0076] Constructing the latent variable subspace of respiratory rhythm component and the latent variable subspace of neural regulation component;

[0077] Through adversarial training, the distribution of respiratory rhythm component subspace is constrained to match the temporal correlation characteristics of respiratory signals;

[0078] In step S2, the RR interval sequence and latent space decomposition are generated as follows:

[0079] The Pan-Tompkins algorithm is used to detect the R wave peak on the original ECG signal x(t) to obtain the peak time sequence And calculate the RR interval: in, represents the (n+1)th R peak moment, Indicates the nth R peak moment, RR n represents the nth RR interval;

[0080] {RR n}After normalization, construct column vector RR=[RR1,RR2,…,RR N ] T , where RR represents the input RR interval sequence column vector and N represents the sequence length; it is fed into the variational autoencoder VAE encoder to obtain the latent space representation: Where z∈R d represents the overall vector of the latent space, represents the encoder mapping, For its parameters;

[0081] Divide z into respiratory rhythm component subspace and neural regulation component subspace according to dimension:

[0082]

[0083] Among them, z r Represents the respiratory rhythm subspace vector with dimension d r , z n Represents the neural modulation subspace vector with dimension d n , d represents the total dimension of the latent space, and R here represents a set of real numbers, which is different from RR;

[0084] Introduce adversarial training and define adversarial loss as:

[0085]

[0086] Among them, L adv represents the adversarial training loss, Represents the real breathing feature sample x r expectations, Represents the encoder output breathing vector z r Expectations, A φ (·) represents the adversarial discriminator mapping, φ is the discriminator parameter, x r Represents the representation of the real breathing signal in the feature domain, z r represents the breathing subspace vector generated by the encoder;

[0087] The encoder and discriminator pass:

[0088] The confrontation process makes z r The distribution of x is gradually aligned r distribution of

[0089] Step S3, performing orthogonal processing on the latent space representation so that the respiratory rhythm component and the neural regulation component are orthogonal in the latent space;

[0090] The orthogonalization process includes:

[0091] Calculate the principal component direction of the respiratory rhythm component in the latent space;

[0092] An orthogonal projection constraint is imposed on the neuromodulatory component so that the angle between it and the principal component direction is no less than 80 degrees;

[0093] In step S3, the steps of orthogonalization of the latent space are as follows:

[0094] Calculate the covariance matrix of the respiratory rhythm subspace vector set, the formula is: C r =Cov(z r ),in, Represents the breathing subspace vector z r The covariance matrix of , Cov(·) is the covariance operator;

[0095] On this basis, by solving the principal component analysis, the main direction vector u of the breathing subspace is obtained:

[0096]

[0097] in, represents the principal component direction vector, |v| represents the Euclidean norm of vector v;

[0098] Perform orthogonal projection on the neuromodulation component vector to obtain the orthogonalized vector:

[0099] z' n =z n -(u T z n )u, where represents the orthogonalized neural modulation vector, u T zn represents the vector inner product;

[0100] Define the orthogonality penalty term:

[0101] L orth =max(0,|u T z n |-|z n |cos(80°)),

[0102] Among them, L orth represents the orthogonality penalty loss, max(·) is the maximum value operator, |z n | represents the Euclidean norm of the neuromodulation vector, cos(80°) represents the cosine value of 80 degrees;

[0103] This term is incorporated into the overall training loss to suppress overlap between subspaces;

[0104] In step S3, an orthogonal projection constraint is applied to the neuromodulation component, specifically:

[0105] The angle between vectors is defined as:

[0106]

[0107] in, represents the principal component direction vector of the breathing subspace, represents the neuromodulatory component vector, |·| represents the Euclidean norm, arccos(·) is the arccosine function, and θ is the angle between vectors;

[0108] The geometric constraint is written as θ≥80°, which is equivalent to:

[0109] Where cos(80°) represents the cosine value of 80°;

[0110] In actual training, a soft constraint penalty term is used:

[0111]

[0112] Among them, L orth is the orthogonality penalty loss, max(·) takes the maximum value operator;

[0113] The threshold setting logic includes the following: In HRV analysis, 80° corresponds to a cosine value of approximately 0.1736, which ensures high subspace separation without excessively weakening neural features. A cosine threshold that is too small will lead to unstable training, while a cosine threshold that is too large will result in insufficient separation. 80° is the empirically established balance point between performance and robustness. For different individuals or application scenarios, 80° can be used as the initial value and fine-tuned based on the performance of the validation set.

[0114] Specifically, this geometric constraint directly limits the minimum angle between the neural modulation component and the main respiratory direction in the latent space, and ensures significant separation between the two in high-dimensional space through the cosine upper bound. Compared with the method of pure projection followed by penalty, the soft constraint introduces an angle threshold, making the filtering effect between subspaces more interpretable. Furthermore, due to the use of a continuous penalty function, the loss is smoothed during training, improving the convergence stability of the model. The 80° threshold not only meets the requirements of HRV signal decoupling, but also retains the subtle fluctuations of neural modulation, providing a purer and more physiologically meaningful RR interval reconstruction result for subsequent frequency domain feature extraction.

[0115] Step S4, reconstructing the RR interval sequence without non-respiratory coupling based on the orthogonalized latent space representation;

[0116] Step S5, calculating frequency domain characteristic parameters based on the reconstructed RR interval sequence and outputting them to the visualization interface of the biofeedback device;

[0117] The calculation steps of frequency domain characteristic parameters include:

[0118] The reconstructed RR interval sequence was corrected in the frequency domain, specifically: the power density value of the region overlapping with the respiratory estimation frequency band was dynamically suppressed in the LF frequency band of 0.04-0.15 Hz;

[0119] In step S5, the frequency domain correction of the reconstructed RR interval sequence is performed as follows:

[0120] Power spectrum estimation is performed using the formula: P(f) = Welch(RR')(f), where RR' represents the reconstructed RR interval column vector, P(f) represents its power spectrum density function, Welch(·) represents the Welch spectrum estimation operator using overlapping windows and averaging, and f represents frequency.

[0121] Define the overlapping area between low frequency band and breathing [f LF,low ,f LF,high ]:

[0122] [f LF,low ,f LF,high ]=[0.04,0.15]Hz,

[0123] F ovl =[f r -δ,f r +δ]∩[f LF,low ,f LF,high ],

[0124] Among them, f LF,low and f LF,high Represent the lower and upper limits of the low-frequency band, f rrepresents the current estimated respiratory rate, and δ represents the half-window width of respiratory rate drift;

[0125] Construct the suppression mask function, expressed as:

[0126] If f∈F ovl , then H(f)=0, otherwise H(f)=1,

[0127] Among them, H(f) is the frequency suppression mask function. When the frequency falls in the overlapping area between the breathing and low frequency bands, F ovl When , it is completely suppressed, otherwise the original power is retained;

[0128] Perform power spectrum correction, the correction formula is: P corr (f)=H(f)P(f), where P corr (f) is the power spectral density function after correction;

[0129] Calculate the low-band correction power P LF :

[0130]

[0131] Among them, P LF represents the low-frequency band power after correction for biofeedback;

[0132] Suppression rules and band selection logic include:

[0133] The low frequency band [0.04, 0.15] Hz can reflect the sympathetic-parasympathetic balance, so it is used as the target power calculation range.

[0134] The respiratory frequency band is usually between 0.1-0.3 Hz and fluctuates with the depth and rate of breathing. r and its drift half-window width δ, which can be set as the standard deviation of the respiratory frequency sliding window Half of the value, or the empirical value is 0.01Hz, dynamically construct the overlapping area F ovl , ensuring that the inhibition rules can adapt to individual breathing changes;

[0135] A hard threshold mask is used to set the power to zero, avoiding the impact of residual respiratory peaks on LF power. This method preserves non-respiratory components while ensuring computational simplicity and real-time performance.

[0136] Specifically, the above mechanism ensures that within the low-frequency band, all power peaks that may be caused by respiratory coupling are dynamically eliminated, thereby providing purer neuromodulation indicators for subsequent biofeedback;

[0137] The biofeedback device data processing method further includes a dynamic adjustment step:

[0138] Monitor the frequency drift of the respiratory signal. When the frequency drift exceeds a preset threshold, it triggers online fine-tuning of the variational autoencoder parameters.

[0139] Online fine-tuning uses an unsupervised loss function and updates the model weights based only on the orthogonality constraint of the latent space of the current input signal;

[0140] The biofeedback device data processing method is deployed on the edge computing device, specifically including:

[0141] Compile the variational autoencoder and orthogonalization steps into a lightweight inference engine; the lightweight inference engine performs model compression through structured pruning and 8-bit integer quantization;

[0142] The RR interval sequence is streamed using memory mapping technology, which performs zero-copy data transfer based on the DMA controller of the target edge device.

[0143] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A biofeedback device data processing method based on artificial intelligence, characterized by: include, Step S1, obtaining the user's electrocardiogram (ECG) signal and respiratory signal, wherein the respiratory signal is generated by a chest and abdomen motion sensor or estimated based on periodic fluctuation characteristics of the ECG signal; Step S2, converting the ECG signal into a RR interval sequence and inputting it into a variational autoencoder for latent space decomposition to obtain a latent space representation including a respiratory rhythm component and a neural regulation component; Step S3, performing orthogonal processing on the latent space representation so that the respiratory rhythm component and the neural regulation component are orthogonal in the latent space; Step S4, reconstructing the RR interval sequence without non-respiratory coupling based on the orthogonalized latent space representation; Step S5: Calculate frequency domain characteristic parameters based on the reconstructed RR interval sequence and output them to the visualization interface of the biofeedback device.

2. The artificial intelligence-based biofeedback device data processing method according to claim 1, characterized in that: The step of generating an estimate of the respiratory signal further comprises: Adaptive sliding window Fourier transform was performed on the RR interval sequence to extract the amplitude change characteristics in the 0.1-0.3 Hz frequency band; A simulated respiratory waveform is generated based on the amplitude variation characteristic as an input substitute for the respiratory signal.

3. The artificial intelligence-based biofeedback device data processing method according to claim 1, characterized in that: The latent space decomposition step of the variational autoencoder includes: Constructing the latent variable subspace of respiratory rhythm component and the latent variable subspace of neural regulation component; The distribution of the respiratory rhythm component subspace is constrained to match the time correlation characteristics of the respiratory signal through adversarial training.

4. The method for processing data of a biofeedback device based on artificial intelligence according to claim 3, wherein: In step S2, the RR interval sequence and latent space decomposition are generated as follows: The Pan-Tompkins algorithm is used to detect the R wave peak on the original ECG signal x(t) to obtain the peak time sequence And calculate the RR interval: in, Indicates the (n+1)th R peak moment, Indicates the nth R peak moment, RR n represents the nth RR interval; {RR n }After normalization, construct column vector RR=[RR1,RR2,…,RR N ] T , where RR represents the input RR interval sequence column vector and N represents the sequence length; it is fed into the variational autoencoder VAE encoder to obtain the latent space representation: Where z∈R d represents the overall vector of the latent space, represents the encoder mapping, For its parameters; Divide z into respiratory rhythm component subspace and neural regulation component subspace according to dimension: Among them, z r Represents the respiratory rhythm subspace vector with dimension d r , z n Represents the neural modulation subspace vector with dimension d n , d represents the total dimension of the latent space, and R here represents the set of real numbers; Introduce adversarial training and define adversarial loss as: Among them, L adv represents the adversarial training loss, Represents the real breathing feature sample x r expectations, Represents the encoder output breathing vector z r Expectations, A φ (·) represents the adversarial discriminator mapping, φ is the discriminator parameter, x r Represents the representation of the real breathing signal in the feature domain, z r represents the breathing subspace vector generated by the encoder; The encoder and discriminator pass: The confrontation process makes z r The distribution of x is gradually aligned r distribution.

5. The method for processing data of a biofeedback device based on artificial intelligence according to claim 1, wherein: The orthogonalization step comprises: Calculate the principal component direction of the respiratory rhythm component in the latent space; An orthogonal projection constraint is imposed on the neuromodulation component so that the angle between it and the principal component direction is not less than 80 degrees.

6. The method for processing data of a biofeedback device based on artificial intelligence according to claim 5, wherein: In step S3, the steps of orthogonalizing the latent space are as follows: Calculate the covariance matrix of the respiratory rhythm subspace vector set, the formula is: C r =Cov(z r ),in, Represents the breathing subspace vector z r The covariance matrix of , Cov(·) is the covariance operator; On this basis, by solving the principal component analysis, the main direction vector u of the breathing subspace is obtained: in, represents the principal component direction vector, |v| represents the Euclidean norm of vector v; Perform orthogonal projection on the neuromodulation component vector to obtain the orthogonalized vector: z' n =z n -(u T z n )u, where represents the orthogonalized neural modulation vector, u T z n represents the vector inner product; Define the orthogonality penalty term: L orth =max(0,|u T z n |-|z n |cos(80°)), Among them, L orth represents the orthogonality penalty loss, max(·) is the maximum value operator, |z n | represents the Euclidean norm of the neuromodulation vector, and cos(80°) represents the cosine value of 80 degrees.

7. The method for processing data of a biofeedback device based on artificial intelligence according to claim 6, wherein: In step S3, an orthogonal projection constraint is applied to the neuromodulation component, specifically: The angle between vectors is defined as: in, represents the principal component direction vector of the breathing subspace, represents the neuromodulatory component vector, |·| represents the Euclidean norm, arccos(·) is the arccosine function, and θ is the angle between vectors; The geometric constraint is written as θ≥80°, which is equivalent to: Where cos(80°) represents the cosine value of 80°; In actual training, a soft constraint penalty term is used: Among them, L orth is the orthogonality penalty loss, and max(·) takes the maximum value operator.

8. The method for processing data of a biofeedback device based on artificial intelligence according to claim 1, wherein: The step of calculating the frequency domain characteristic parameters includes: The reconstructed RR interval sequence was corrected in the frequency domain, specifically by dynamically suppressing the power density value in the region overlapping with the respiratory estimation frequency band within the LF frequency band of 0.04-0.15 Hz.

9. The method for processing data of a biofeedback device based on artificial intelligence according to claim 8, wherein: Further including the dynamic adjustment step: Monitoring the frequency drift of the respiratory signal, and triggering online fine-tuning of the parameters of the variational autoencoder when the frequency drift exceeds a preset threshold; The online fine-tuning adopts an unsupervised loss function and updates the model weights only based on the orthogonality constraint of the latent space of the current input signal.

10. The method for processing data of a biofeedback device based on artificial intelligence according to claim 9, wherein: This method is deployed on edge computing devices and specifically includes: Compiling the variational autoencoder and the orthogonalization processing step into a lightweight inference engine; the lightweight inference engine performs model compression through structured pruning and 8-bit integer quantization; The RR interval sequence is streamed using a memory mapping technique that performs zero-copy data transfer based on a DMA controller of a target edge device.

Images