Sleep environment feedback control data processing method based on cognitive behavior therapy logic

Abstract

The invention relates to the technical field of data processing and intelligent environment control, in particular to a sleep environment feedback control data processing method based on cognitive behavior therapy logic, which comprises the following steps: a reference model construction step: acquiring a non-contact multi-mode time sequence signal, and reconstructing a reference signal balance curve to construct a target signal manifold model; a simulation flow generation step of converting the preset interference characteristics into signal modulation operators and injecting the signal modulation operators into the model to generate a simulation signal flow; a double-track difference calculation step: executing double-track difference operation to respectively extract residual features representing system deviation and theoretical interference; a similarity judgment and control step: calculating the topological similarity of the residual features in a time-frequency domain, judging a deviation dominant type, generating a corresponding control instruction, and driving an execution mechanism to act; according to the method, psychological anxiety and environmental discomfort are effectively distinguished by using a double-track differential mechanism, and accurate attribution analysis of sleep interference sources is realized.

Description

Technical Field

[0001] This invention relates to the fields of data processing and intelligent environmental control technology, specifically a sleep environment feedback control data processing method based on cognitive behavioral therapy logic. Background Technology

[0002] Existing intelligent sleep monitoring and feedback control systems typically include a non-contact sensor array placed inside the mattress or pillow, as well as an actuator capable of adjusting temperature, support, or height. In related technologies, to achieve intervention and control of the sleep process, triggering logic based on physiological signal thresholds is mainly adopted. For example, by collecting the user's body movement amplitude or heart rate data, when the detected signal intensity exceeds the preset resting threshold, the system determines that the user is in an agitated or light sleep-wake state, and then directly generates control commands to adjust the air conditioner temperature or change the mattress firmness parameters, attempting to soothe the user by improving the physical environment. However, when faced with complex sleep disorder scenarios, the control method that relies solely on signal thresholds has significant defects. On the one hand, the existing technical logic cannot effectively distinguish whether the agitation signal generated by the user is a high arousal state caused by psychological anxiety or a physiological adjustment caused by physical discomfort, resulting in ambiguous attribution. On the other hand, when the user is in a wakeful state caused by rumination or anxiety, if the system mistakenly determines it as environmental discomfort and performs physical adjustments, such as lowering the temperature or mechanical vibration, it will not only fail to solve the user's psychological insomnia problem, but may also aggravate the user's agitation due to incorrect disturbance. Therefore, the sleep environment feedback control method in related technologies struggles to simultaneously achieve accurate identification of interference source types and effective closed-loop control. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a sleep environment feedback control data processing method based on cognitive behavioral therapy logic. Specifically, the technical solution of this invention includes: Collect non-contact multimodal time-series signals within the target monitoring area and obtain preset system reference parameters; Based on the system baseline parameters, the baseline signal equilibrium curve over time is reconstructed, and a target signal manifold model representing the target state is constructed. The preset interference characteristics are transformed into signal modulation operators, and the signal modulation operators are injected into the target signal manifold model to generate a simulated signal flow containing theoretical interference characteristics. Dual-track differential operation is performed on non-contact multimodal time-series signals, target signal manifold models, and simulated signal streams to extract the first residual feature characterizing the current system signal deviation and the second residual feature characterizing the purely theoretical interference signal, respectively. Calculate the topological similarity between the first residual feature and the second residual feature in the time-frequency domain feature space; Based on topological similarity, the dominant type of signal deviation is determined, and corresponding first-type or second-type control commands are generated according to the dominant type to drive the actuator to act, forming an environmental control closed loop.

[0004] Preferably, the non-contact multimodal time-series signal acquisition within the target monitoring area includes: Vibration signals and periodic waveform data of the target area are collected using piezoelectric ceramic sensors or fiber optic sensor arrays. Dynamic micro-variation signals in the target area are acquired by using a pressure distribution matrix sampled at high frequency. The vibration signal, periodic waveform data, and dynamic micro-variable signal are cleaned and synchronized to generate a basic input data stream for subsequent processing.

[0005] Preferably, constructing a target signal manifold model representing the target state includes: Within the target time window, the frequency of the dynamic micro-variable signal under the baseline state exhibits an exponential decay trend. The high-frequency components of the signal rhythm variability under the set reference state exhibit a preset rising slope. By combining the exponential decay trend and the rising slope, a baseline signal model is generated, which serves as the target signal manifold model.

[0006] Preferably, the process involves converting preset interference characteristics into signal modulation operators and injecting these operators into the target signal manifold model to generate a simulated signal flow containing theoretical interference characteristics, including: Construct impulse noise that conforms to non-periodic abrupt changes as a first-class feature injection operator; A filter simulating a specific low-frequency suppression effect is constructed as a second-type feature injection operator; The first type of feature injection operator and the second type of feature injection operator are superimposed on the target signal manifold model to generate a simulated signal flow containing theoretical interference features.

[0007] Preferably, a dual-track differential operation is performed on the non-contact multimodal time-series signal, the target signal manifold model, and the simulated signal stream to extract the first residual feature and the second residual feature, respectively: The first difference between the basic input data stream and the target signal manifold model is calculated to obtain the first residual feature; Calculate the second difference between the simulated signal flow and the target signal manifold model to obtain the second residual characteristic; The first residual feature contains mixed information of environmental noise, physical state changes, and theoretical interference, while the second residual feature contains only the signal features caused by theoretical interference.

[0008] Preferably, calculating the topological similarity between the first residual feature and the second residual feature in the time-frequency domain feature space includes: The first residual feature and the second residual feature are transformed in the time and frequency domain respectively, and the transformed feature coefficients are extracted. The distance between feature coefficient sequences can be calculated using a dynamic regularization algorithm, or the angle between feature coefficients can be calculated using a vector similarity algorithm, to obtain quantified topological similarity values.

[0009] Preferably, determining the dominant type of signal deviation based on topological similarity and generating corresponding control commands based on the dominant type includes: The topological similarity value is compared with a preset decision threshold; If the topological similarity value is greater than the decision threshold, it is determined that the current signal deviation is mainly dominated by theoretical interference features; In response to this determination, a first type of control command is generated; the first type of control command is used to drive the actuator to initiate rhythmic guidance or quickly adjust the first environmental parameters.

[0010] Preferably, determining the dominant type of signal deviation based on topological similarity and generating corresponding control commands based on the dominant type further includes: If the topological similarity value is less than the decision threshold, it is determined that the current signal deviation is mainly caused by environmental noise or unsuitable physical conditions. In response to this determination, a second type of control command is generated; the second type of control command is used to drive the actuator to adjust the physical support parameters or slowly adjust the second environmental parameters.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention introduces a simulated signal stream as an intermediate reference variable and performs dual-track differential operations to extract the first residual feature representing the current system signal deviation and the second residual feature representing the purely theoretical interference signal, and then calculates the topological similarity between the two in the time-frequency domain feature space. This scheme effectively solves the technical pain point of existing technologies that cannot distinguish between physiological wakefulness and psychological hyperarousal by relying solely on body movement thresholds. By quantitatively comparing the topological morphology of real-time signals and theoretical anxiety models, the system can accurately identify the driving source behind user agitation, thereby avoiding erroneous physical environment adjustment when the user is awake due to psychological anxiety, and realizing accurate attribution analysis of sleep disturbance sources. 2. This invention, by fusing vibration signals from piezoelectric or fiber optic sensors with dynamic micro-variation signals from a high-frequency pressure distribution matrix, can simultaneously capture a user's endogenous physiological rhythms and exogenous limb behaviors, solving the problem of insufficient information dimension from a single sensor. By employing a dual synchronization strategy of coarse alignment and fine alignment based on Hilbert transform energy envelope and cross-correlation function, and utilizing cubic spline interpolation and wavelet thresholding for denoising, this scheme effectively eliminates clock drift and signal source aliasing between different sensors, ensuring strict alignment and smoothness of multidimensional data on the time axis. This provides a high signal-to-noise ratio and dimensionally consistent basic input data stream for subsequent complex algorithm analysis. 3. This invention reconstructs the baseline signal equilibrium curve over time based on system baseline parameters, and constructs a target signal manifold model by combining the kinetic density exponential decay trend and the vagal nerve activity rise slope. This scheme establishes a gold standard reference system that eliminates pathological features and environmental interference, enabling the assessment of sleep state to have a dynamically changing mathematical benchmark. By transforming the conditioned arousal and rumination characteristics in cognitive behavioral therapy theory into mathematical signal modulation operators, such as impulse noise operators and low-frequency suppression filters, and injecting these operators into the ideal model to generate a simulated signal stream, this scheme successfully transforms the abstract psychological anxiety state into a computer-processable digital waveform, enabling the system to quantify the anxiety level by calculating the distance or angle of the residual features, overcoming the shortcomings of traditional methods in quantifying psychological indicators. 4. This invention determines the dominant type of signal deviation based on topological similarity and generates differentiated control commands accordingly. When it is determined to be psychological hyperarousal, a first type of control command is generated to block rumination. When it is determined to be physical environmental interference, a second type of control command is generated to improve comfort. This solution achieves precise closed-loop control that conforms to the logic of cognitive behavioral therapy, breaking the traditional single comfort adjustment mode of devices. By taking proactive intervention when psychological factors are identified and adaptive compensation when environmental factors are identified, it minimizes unintended disturbances to users and ensures the pertinence and effectiveness of intervention measures. Attached Figure Description

[0012] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0014] Example 1: Please see Figure 1A sleep environment feedback control data processing method based on cognitive behavioral therapy logic includes: collecting non-contact multimodal time-series signals within the target monitoring area and obtaining preset system baseline parameters; Based on the system baseline parameters, the baseline signal equilibrium curve over time is reconstructed, and a target signal manifold model representing the target state is constructed. The preset interference characteristics are transformed into signal modulation operators, and the signal modulation operators are injected into the target signal manifold model to generate a simulated signal flow containing theoretical interference characteristics. Dual-track differential operation is performed on non-contact multimodal time-series signals, target signal manifold models, and simulated signal streams to extract the first residual feature characterizing the current system signal deviation and the second residual feature characterizing the purely theoretical interference signal, respectively. Calculate the topological similarity between the first residual feature and the second residual feature in the time-frequency domain feature space; determine the dominant type of signal deviation based on the topological similarity, and generate the corresponding first type or second type of control command according to the dominant type to drive the actuator to act and form an environmental control closed loop.

[0015] This embodiment details the execution logic of the above method. The method mainly runs in the central processing unit or cloud server of the smart mattress system, aiming to solve the technical pain point that the existing technology cannot distinguish between physiological wakefulness and psychological hyperarousal. The system collects non-contact multimodal time-series signals in the target monitoring area in real time through a sensor array deployed inside the mattress. These signals include cardiac impaction, respiratory waves and body movement signals. At the same time, it obtains preset system reference parameters, which are derived from the sleep window set by the user and the standard autonomic nervous balance index based on age and gender. Based on this, the system does not directly use the original signal for analysis. Instead, it reconstructs a baseline signal equilibrium curve in the mathematical topological space based on the system's baseline parameters. This curve simulates an ideal process in which a healthy sleeper experiences a decrease in sympathetic nerve activity, an increase in parasympathetic nerve activity, and an exponential decay in body movement frequency during the process of falling asleep. This process constructs a target signal manifold model M0 that represents the target state. This model eliminates all pathological features and environmental interference and serves as the gold standard reference system. To identify psychogenic insomnia, the system transforms pre-defined interference features—namely, conditioned arousal and rumination features derived from CBT-I theory—into a mathematical signal modulation operator. This operator is then injected into an ideal M0 to generate a simulated signal stream M representing a theoretically severe psychological anxiety experience. s The system processes real-time acquired non-contact multimodal time-series signals S. r Perform a two-track difference operation: the first track calculates S. r The difference from M0 is used to extract the first residual feature R1, representing the degree to which the user's current state deviates from perfect sleep; the second track calculates M.s The difference from M0 is used to extract the second residual feature R2, which represents the theoretical signal feature caused purely by psychological anxiety; The system calculates the topological similarity between R1 and R2 in the time-frequency domain feature space; based on this similarity, the system determines the dominant type of signal deviation: if the similarity is high, it is determined to be psychological high arousal, and a first type of control command such as CBT-I guided intervention is generated; if the similarity is low, it is determined to be physical or environmental interference, and a second type of control command such as adjusting the firmness of the air conditioner or mattress is generated to drive the actuator to act, forming an environmental control closed loop. In this embodiment, under the intelligent sleep monitoring scenario, a simulated signal stream is introduced as an intermediate reference variable to achieve accurate attribution analysis of sleep disturbance sources. Compared with the traditional technology that triggers control based solely on body movement thresholds, this solution utilizes a dual-track differential mechanism to effectively distinguish whether the user's wakefulness is due to psychological anxiety or agitation due to environmental discomfort. This avoids disturbances caused by incorrect temperature adjustments when the user is awake due to anxiety, and achieves accurate closed-loop control that conforms to CBT-I medical logic.

[0016] Example 2: The acquisition of non-contact multimodal time-series signals within the target monitoring area includes: acquiring vibration signals and periodic waveform data of the target area through piezoelectric ceramic sensors or fiber optic sensor arrays; acquiring dynamic micro-variable signals of the target area through a high-frequency sampled pressure distribution matrix; and cleaning and synchronizing the vibration signals, periodic waveform data, and dynamic micro-variable signals to generate a basic input data stream for subsequent processing.

[0017] This embodiment further defines the hardware architecture and processing logic of the signal acquisition steps, aiming to solve the problem of signal source aliasing in non-contact monitoring; the system acquires high-sensitivity vibration signals and periodic waveform data through piezoelectric sensors or fiber optic grating sensor arrays laid on the mattress surface, wherein the vibration signals include micro-vibrations caused by heartbeats, and the periodic waveform data corresponds to respiratory waves; Simultaneously, the system utilizes a high-frequency pressure distribution matrix composed of capacitive or resistive pressure sensor arrays to acquire dynamic micro-variation signals in the target area at a sampling rate exceeding 50Hz. These signals reflect subtle shifts in the user's center of gravity. To eliminate clock drift between different sensors, the system performs rigorous data cleaning and synchronization alignment operations: Let the piezoelectric signal sequence be X... p The pressure matrix signal sequence is X m ; Considering that the sampling rate of the pressure matrix may be insufficient to capture a clear R-wave peak in some low signal-to-noise ratio scenarios, the system adopts a dual synchronization strategy of coarse alignment + fine alignment: calculating X p With X mThe global energy envelope, here, to clarify the computer implementation logic of the energy envelope, is defined for the vibration signal X. p The system uses the Hilbert transform to extract the magnitude of the analytic signal: Where H[⋅] is the Hilbert transform operator; for the pressure matrix signal X m The system calculates the Frobenius norm of the matrix in each frame as the instantaneous energy value: ; The above formula uses the inter-frame difference method or the mean removal method to calculate the instantaneous dynamic energy, in order to eliminate the DC component generated by static body weight and ensure the extracted E m (t) only includes dynamic changes caused by body motion, thus possessing the characteristics of vibration signal E p (t) The physical basis for performing cross-correlation calculations; Based on the calculated E p With E m E with a higher sampling rate p Downsampling or resampling both to a uniform frequency, such as 50Hz, ensures consistent discrete-time series lengths; the maximum correlation time of their envelopes is calculated using the cross-correlation function (CCF). This is used as a rough time deviation compensation; based on this, X is then used... p Using the peak time of the R-wave as the reference anchor point, the cubic spline interpolation algorithm is used to interpolate the X-wave. m Timestamps aligned to X p The timeline is optimized to ensure the smoothness and continuity of the time series; wavelet thresholding is used for denoising, specifically employing the Daubechies4 (db4) wavelet basis for three-level decomposition and applying a soft thresholding function to process the wavelet coefficients; to ensure the reproducibility of the algorithm, the soft thresholding function is defined as: ;

[0018] Where w represents the original wavelet coefficients; the threshold λ is calculated using a general threshold formula: ; Where N is the signal length and σ is the noise standard deviation. n The MAD (mean squared absolute difference) of the first-level decomposition coefficients is estimated to be: Here, MAD is defined as: ; Given that the mean of the wavelet coefficients is approximately 0, it can be simplified to The constant 0.6745 here represents the 0.75 quantile of the standard normal distribution, used to convert the absolute median into an unbiased estimate of the standard deviation against Gaussian white noise. This specific parameter configuration removes power frequency interference, ultimately generating the base input data stream for subsequent processing. In this embodiment, under the scenario of multimodal data fusion, by fusing one-dimensional time-series waveforms and two-dimensional spatial distribution data, it is possible to simultaneously capture the user's endogenous physiological rhythms and exogenous physical behaviors; this multi-dimensional acquisition strategy provides a complete data foundation for subsequent differentiation between psychological anxiety and physical discomfort.

[0019] Example 3: The construction of the target signal manifold model representing the target state includes: setting the frequency of the dynamic micro-variable signal under the reference state to exhibit an exponential decay trend within the target time window; setting the high-frequency component of the signal rhythm variability under the reference state to exhibit a preset rising slope; and combining the exponential decay trend and the rising slope to generate a reference signal model as the target signal manifold model.

[0020] This embodiment details the construction logic of the target signal manifold model. This model is not a static numerical model, but a set of functions that dynamically change with the sleep onset time t. Based on the sleep drive theory in CBT-I, the system is set to assume that the frequency of the dynamic micro-variable signal under the baseline state exhibits an exponential decay trend within the target time window. Based on the equivalence principle in physics, the signal frequency mentioned in the embodiment is represented as kinetic energy density in the mathematical model. That is, it is assumed that the frequency of micro-motion events per unit time is proportional to the released physical kinetic energy, thus introducing an ideal kinetic energy decay function E(t) for characterization, as shown in the following formula: ; E0 is derived from the average body motion amplitude in the 5 minutes prior to the user's bedtime, physically representing the initial body motion energy, measured in gravitational acceleration g. The specific calculation method is as follows: Accelerometer data a(t) from the 5 minutes prior to bedtime is extracted and calculated using a sliding window root mean square algorithm. ; To accurately reflect the effective value of body kinetic energy; to prevent subsequent logarithmic calculations from failing or the decay constant from being abnormal due to the user initially being in a static state (E0≈ϵ), the system applies a lower bound constraint to E0 before calculation: This ensures the mathematical existence of the decay trend; β is derived from statistical data on sleep latency in healthy individuals. To implement the variable tracking required in Rule 1, the mapping relationship of the age parameter is clarified here: the system calculates the standard sleep latency T based on the user-inputted age (Age), which ranges from 18 to 80 years old. lat The specific calculation formula is defined as follows: ; To address the issue of unclear basis for formula construction, it is clarified here: the constant 1200 represents the baseline physiological sleep latency for people aged 25 and under, i.e., 20 minutes × 60 seconds, with the time dimension of seconds; the constant 30 is the age-related decline coefficient, with the dimension of seconds / year, representing the statistical trend that the latency increases by an average of 30 seconds for every additional year of age; this formula and its coefficients are derived from linear regression analysis of sleep data of healthy individuals aged 20-60 years in the system database. The system database specifically refers to a collection of sleep data from 1500 healthy adults annotated with the gold standard of polysomnography (PSG), such as the healthy control group subset of the SleepHeartHealthStudy, excluding apnea samples with AHI>5. The regression analysis sample size N=1500, the correlation coefficient R²>0.8, and a clear physical relationship between numerical values ​​and time units was established. The specific calculation logic is as follows: based on the user's specific T lat And assume that at t=T lat The kinetic energy decays to a preset convergence threshold E close to the noise floor ϵ. conv ,set up: ; Where, δ margin Let β be the preset safety margin coefficient, with a value of 0.1; then β passes through the boundary condition equation: ; The analysis yielded the following calculation formula: ; This ensures that the decay constant has clear physical boundary constraints and is related to the user's physiological characteristics; t: derived from the system timer, its physical meaning is the normalized sleep time, and the unit is seconds; ϵ originates from sensor noise floor measurement, its physical meaning is basic physiological micro-motion noise floor, and its unit is gravitational acceleration g; To characterize the ideal activation process of the parasympathetic nervous system, the system is set to assume that the high-frequency components of the signal rhythm variability under the baseline state exhibit a preset rising slope, and the ideal vagal nerve activity function H(t) is defined as follows: ; Here, α is derived from preset physiological model parameters. To implement the variable tracking required in the rules, the mapping relationship between gender and autonomic balance index is explicitly defined: the system retrieves gender data from the user profile and obtains the corresponding standard autonomic balance index (ABI). std The specific mapping logic is as follows: if the gender is male, set ABI. std =1.0; If the gender is female, set ABI. std=1.2; This value is determined based on the sex difference norm of heart rate variability (HRV). Specifically, it refers to the 24-hour HRV norm study by Umetani et al. The average SDNN and RMSSD in women aged 20-60 years are about 20% higher than those in men. Therefore, a coefficient of 1.2 is set to reflect the relatively higher vagal tone in women at rest. The specific calculation logic is as follows: obtain the user's resting heart rate reserve (HRR) and set it at the ideal sleep latency (T). lat The process of vagal tone elevation from wakefulness to light sleep is completed, with target increments ΔH, HRR, and ABI. std If they are directly proportional, then: ; in, κ is the dimension conversion coefficient; to address the issue of insufficient parameter sources, the method for obtaining κ is explained here: select healthy sleep samples from a subset of the NormalSinusRhythmDatabase in MIT-BIH, a total of 18 long-term ECG records, and extract HRRbpm and HF power increment ΔHms² during the sleep process; The specific extraction steps are as follows: resample the RR interval sequence of each sample at 4Hz, calculate the power spectral density of the 5-minute sliding window, and integrate to obtain the energy in the 0.15-0.4Hz frequency band; perform linear regression fitting on the extracted effective data pairs. Thus, the value of κ is determined to be 0.5, with dimensions m / s. 2 / bpm is directly derived from this physical division relationship, representing the vagus nerve power gain per unit of heart rate reserve; H0 comes from the initial measurement value, and its physical meaning is the initial high-frequency component power, with the unit being milliseconds squared (ms). 2 ); The system combines the exponential decay trend E(t) and the rising slope H(t) to generate a reference signal model in a multidimensional vector space, which serves as the target signal manifold model. To ensure that this manifold model M0 is strictly aligned in dimension with the multimodal data collected in Example 2 for subsequent differencing, the system performs specific channel mapping: mapping E(t) to the kinetic energy amplitude dimension of the corresponding pressure distribution matrix signal, and mapping H(t) to the frequency domain variability dimension of the corresponding vibration signal, thereby constructing a time-series state vector consistent with the number of feature channels of the basic input data stream. This ensures the mathematical feasibility of subsequent dual-track difference operations.

[0021] Example 4: The process involves converting preset interference features into signal modulation operators and injecting these operators into the target signal manifold model to generate a simulated signal flow containing theoretical interference features. This includes: constructing impulse noise that conforms to non-periodic abrupt changes as a first-type feature injection operator; constructing a filter that simulates a specific low-frequency suppression effect as a second-type feature injection operator; and superimposing the first-type and second-type feature injection operators onto the target signal manifold model to generate a simulated signal flow containing theoretical interference features.

[0022] This embodiment details how to transform an abstract state of anxiety into a computer-processable signal modulation operator. To simulate somatic anxiety in a high-arousal state as defined in CBT-I, the system constructs a pulse noise N(t) conforming to non-periodic mutations as a first-class feature injection operator. This operator simulates the sudden, irregular, high-amplitude body movements that occur in insomnia patients during their calm period, mathematically expressed as a set of randomly distributed Dirac function convolutions. ; in, The Dirac function is used here for mathematical modeling. Instantaneous impact events occurring at a specific moment; in the discrete implementation of digital processors, to solve the problem of amplitude singularity of the continuous Dirac function at sampling points, Specifically, it is instantiated as a discrete unit sample sequence δ[n], that is, when the sampling index n corresponds to time... When δ[n] = 1, otherwise it is 0; at this time A k It is directly used as the instantaneous acceleration amplitude superposition at discrete sampling points, thereby ensuring that the generated noise sequence is strictly consistent with the acceleration unit g in the target signal manifold model in terms of physical dimensions; A k Derived from a random number generator, its physical meaning is to simulate the random amplitude of body movement, with the unit being gravitational acceleration g; to simulate the characteristics of involuntary myoclonus, A k It follows a uniform distribution U(0.1,0.8) on the interval [0.1g,0.8g], which covers the typical amplitude range from slight twitching to significant limb movement; Derived from the Poisson distribution model, the rate parameter λ of the Poisson process is set to 5 times per minute. This parameter is determined based on the average body movement statistics during the sleep latency of clinical anxiety patients. Specifically, it is based on the statistical data of the average incidence of myoclonus in the N1 phase of psychophysiological insomnia patients diagnosed under the ICSD-3 criteria. This feature maps the somatization response caused by hypervigilance described in CBT-I. K is derived from the simulation duration setting and its physical meaning is the total number of pulses. Meanwhile, to simulate the elevated cortisol levels and sustained sympathetic nerve excitation caused by anxiety, i.e., the inhibition of the high-frequency (HF) component of the parasympathetic nervous system, a filter simulating a specific low-frequency inhibition effect is constructed as a second-type feature injection operator, specifically manifested as a low-frequency gain filter G. lf The filter is designed as a second-order Butterworth low-pass filter with a cutoff frequency set at 0.15Hz to ensure effective suppression of high-frequency energy in the 0.15-0.4Hz range. This frequency range strictly corresponds to the HRV high-frequency HF band defined by the European Society of Cardiology (ESC) and the National Association for Pacing and Electrophysiology (NASPE) of North America, which reflects vagal nerve activity. The mathematical operations used to suppress it clearly map the phenomenon of vagal nerve tension suppression caused by elevated cortisol levels due to rumination in the CBT-I theory. The system superimposes the first type of feature injection operator and the second type of feature injection operator onto the target signal manifold model M0(t) to generate a simulated signal flow M containing theoretical interference features. s (t); In this step, in order to solve the dimension matching problem between the signal layer operator and the feature layer model, the superposition and convolution operations are specifically implemented through feature domain mapping: For the kinetic energy amplitude dimension E(t): the system calculates the root mean square (RMS) envelope of the impulse noise N(t) within the sliding window. Here, to clarify the calculation boundary, the sliding window width W is set. rms =2.0 seconds, sliding step size Δt step =0.1 seconds, generating the energy perturbation term ΔE(t), the specific calculation formula is as follows: ; This step performs energy smoothing on the discrete impulse noise sequence N(t) along the time axis, transforming it into a continuous energy envelope consistent with the dimension of the target signal manifold model E(t); and then injects it in an additive manner: E s (t)=E(t)+ΔE(t) For the vagal nerve activity dimension H(t) (the second component of M0): the system calculates the filter G. lf The theoretical power attenuation ratio γ (γ∈[0,1]) for the high-frequency band (0.15-0.4Hz); the specific formula for calculating γ is: ; Among them, f l =0.15Hz, f h =0.4Hz; Here, to clarify the mathematical definition of the integral kernel, H G (f) refers to the frequency response function of a second-order Butterworth filter, whose squared magnitude |H G (f)| 2 The specific analytical expression for power gain is defined as follows: ; Where the filter order n=2, and the cutoff frequency f c The frequency is set to 0.15Hz; this formula clarifies the energy transfer characteristics of the filter in the frequency domain, and the square term is used to maintain physical consistency with the power dimension of H(t); this coefficient characterizes the relative suppression effect of low-frequency component enhancement on high-frequency components, which the system injects in the form of multiplicative modulation: ; Through the channel-level mapping calculations described above, the system successfully constructed a simulated signal flow vector isomorphic to M0(t): .

[0023] Example 5: A dual-track differential operation is performed on the non-contact multimodal time-series signal, the target signal manifold model, and the simulated signal stream to extract the first residual feature and the second residual feature, respectively. The first residual feature is obtained by calculating the first difference between the basic input data stream and the target signal manifold model; the second residual feature is obtained by calculating the second difference between the simulated signal stream and the target signal manifold model. The first residual feature contains mixed information of environmental noise, physical state changes, and theoretical interference, while the second residual feature contains only the signal features caused by theoretical interference.

[0024] This embodiment details the specific implementation logic of the dual-track differential operation; in order to ensure the basic input data stream S r The dimension of the constructed two-dimensional target signal manifold model M0(t) in the vector space is strictly matched with that of the target signal manifold model M0(t). The system then performs the preliminary feature channel construction step: For the first feature channel, the system uses the differential energy integration method to calculate the real-time kinetic energy amplitude E from the cleaned pressure distribution matrix P(x,y,t). real (t), the specific formula is: ; Where X and Y are the matrix dimensions, Δt is the sampling interval, and η is the preset normalization coefficient; the specific formula for calculating η is: ; Among them, M kg Derived from user profile data, its physical meaning is the user's weight, in kg; C FSR P represents the theoretical resultant force value corresponding to the full scale of the pressure sensor array, in N; max This represents the theoretical full-scale value of the pressure sensor's analog-to-digital converter (ADC); g const The standard gravitational acceleration constant is taken as 9.8 m / s². 2 This formula, based on Newton's second law a=F / m, maps the dimensionless ADC sample value to a kinetic energy amplitude in units of gravitational acceleration g. To satisfy the traceability rule for variable definition, P is explicitly defined here. max This represents the theoretical full-scale value of the pressure sensor's analog-to-digital converter (ADC), for example, P in a 12-bit sampling system. max =4095, this coefficient ensures that the original sensor code value is correctly mapped to the physical unit of gravitational acceleration g; for the second feature channel, the system uses the time-frequency domain integration method to calculate the high-frequency component H of real-time rhythm variability from the vibration signal V(t). real (t): Perform sliding window peak detection on V(t) to extract the heartbeat time sequence t. k Construct RR interval sequences ; Using cubic spline interpolation to transform RR k The time series RR(t) is resampled to a uniform time series of 4 Hz; a short-time Fourier transform (STFT) is performed on RR(t) with a window length of 30 seconds and a sliding step size of 0.25 seconds to generate a high-resolution time-frequency matrix with strictly aligned time axes, and the power spectral density PSD(f,t) is calculated; the high-frequency power is obtained by integration. ; This instantiates the basic input data stream into a two-dimensional measured state vector isomorphic to M0(t). To eliminate the interference of high-frequency measurement noise from the micro-motion sensor on trend comparison and prevent noise variance from dominating residual calculation, the system... r (t) The vector sequence is subjected to moving average filtering, with a window length set to W. ma =10 seconds, generate smoothed observation vector Used for subsequent calculations; Based on this, the system calculates the smoothed base input data stream. The first difference between the target signal manifold model M0(t) and the target signal is calculated using a modified standardized Euclidean distance to obtain the first residual feature R1(t) in order to eliminate the influence of dimensions and prevent the denominator from being zero; σ hist The specific calculation method is as follows: Extract the user's historical sleep data from the past 7-14 days, remove awake period data, and then calculate the global standard deviation of each dimension of the signal. The specific removal strategy is as follows: Call the user's historical sleep stage report stored in the system database, and only select the data from the time period marked as N2 or N3 deep sleep stage to participate in the standard deviation calculation, so as to exclude high variance data from the awake period from affecting σ. hist Weight contamination; If the target user is a first-time user, then the general standard deviation norm σ is used. pop Substitution is performed; to satisfy the threshold and parameter source rules, σ is explicitly defined here. pop Specific values ​​and sources: σ pop It is a two-dimensional vector The data were derived from a statistical analysis of sleep data from 5,000 healthy subjects. To ensure the verifiability of the parameter sources, the dataset was taken from the publicly available Sleep Heart Health Study SHHS-1 baseline database. The selection criteria were non-medication subjects with an AHI < 5 and a sleep efficiency > 85%. Specifically, the data was set as follows: The average variability corresponding to the kinetic energy amplitude dimension. This corresponds to the average variability of the high-frequency component dimension; this parameter is embedded in the system firmware to ensure that the algorithm can still calculate the residual characteristics based on the population average baseline when new users lack historical N2 / N3 data; the formula is expressed as: ; Where i represents the vector dimension index, and ξ is a preset minimum regularization constant, with a value of 1e. -6 The selection of this value is based on the following: it is set to be less than the minimum resolution of the system sensor (1e). -5 g) A value of an order of magnitude that can prevent the computational singularity of zero denominator without significantly interfering with the physical residual value. System calculation simulation signal flow M s The second difference between R1(t) and the target signal manifold model M0(t) yields the second residual feature R2(t). To ensure mathematical comparability between R1 and R2 in subsequent steps and to prevent similarity calculation failure due to inconsistent measurement scales, the system must use a normalized weight σ that is completely consistent with R1. hist To calculate R², the explicit formula is as follows: ; For users' historical standard deviation σ hist Explanation of the rationale for normalizing the purely theoretical deviation R2: Since subsequent steps require calculating the topological similarity between R1 (measured deviation) and R2 (theoretical deviation), it is required that both must be in the same metric space; R1 has been normalized by σ. hist Normalized to the deviation factor relative to the user's reference noise, i.e., the signal-to-noise ratio, therefore R² must also be divided by the same σ. hist This is converted into the expected deviation factor of the theoretical interference relative to the user's reference noise; Without this unified normalization, the numerical magnitudes of the two will be completely unbalanced. For example, R1 is a dimensionless value and R2 is a physical quantity value, which will cause the similarity calculation to fail. This step ensures that the residual sequence generated afterward reflects the relative deviation rather than the absolute numerical difference in topological form by mapping the theoretical interference to the user-specific historical variation space.

[0025] Example 6: Calculating the topological similarity between the first residual feature and the second residual feature in the time-frequency domain feature space includes: performing time-frequency domain transformation on the first residual feature and the second residual feature respectively, and extracting the transformed feature coefficients; using the dynamic warping algorithm to calculate the distance between the feature coefficient sequences, or using the vector similarity algorithm to calculate the angle between the feature coefficients, to obtain the quantized topological similarity value.

[0026] This embodiment details how to quantify the similarity between two residual features; given the potential overlap of anxiety features in the purely time domain, the system defines a similarity of length T. obs A rolling observation window of 300 seconds is used, and the sequences of the first residual feature R1 and the second residual feature R2 within this window are extracted and subjected to continuous wavelet transforms. Here, the complex Morlet wavelet is selected as the basis function, and the center frequency is set to f. c =1.0Hz, bandwidth parameter f b =1.5; Extract the transformed feature coefficients C1(ω,t) and C2(ω,t); given that the output of the complex Morlet wavelet transform is a complex matrix, the system specifically performs modulus operations to obtain the real-valued wavelet energy spectrum; Before performing similarity calculations, in order to eliminate the influence of differences in absolute signal amplitude on topological shape judgment, that is, to avoid distance deviations caused by different signal gains and to ensure that shape rather than intensity is being measured, the system performs Z-Score normalization on matrices C1 and C2 respectively: ; Where, μ C ,σ C These are the global mean and standard deviation of the corresponding matrices, respectively. To quantify the morphological similarity between the two in terms of evolutionary trends, the system processes the standardized feature matrix according to the selected algorithm: if the vector similarity algorithm is selected, the system performs a global flattening operation to reshape the matrix into a one-dimensional feature vector and calculates the cosine similarity. If the Dynamic Warping (DTW) algorithm is selected, the system will generate the frequency vector u corresponding to each time t. t As nodes, the sequence distance is calculated using the dynamic regularization algorithm; in the recursive calculation step of the dynamic regularization algorithm, the i-th feature vector u in the reference sequence is defined. i With the j-th feature vector v in the real-time sequence j Local distance cost function d(u) i ,v j Since the feature matrix has been Z-score normalized in the previous steps, Euclidean distance is used here as the metric to capture the absolute differences in spectral energy distribution: ; Where m is the frequency dimension index of the feature vector, and F is the total number of frequency points; the cumulative distance matrix D of DTW is calculated using a recursive formula: ; To prevent ill-conditioned distortion of the time axis, such as mismatching features at the end of the model to the beginning of the real-time signal, the DTW algorithm introduces the Sakoe-Chiba banding constraint to restrict the regularized path w(k) = (i k ,j k ) satisfies |i k -j k | <R band Where the constraint radius R band Set to 10% of the time series length; use a Gaussian kernel function to obtain the quantized topological similarity value: ; Here, to avoid the filter attenuation coefficient in Example 4 A symbol definition conflict has occurred, so the kernel function parameter has been renamed. To resolve the issue of the parameter's origin, the system sets it as follows: ; Among them, D ref The preset distance reference constant is T. obs ×0.5; the 0.5 coefficient here is based on the fact that the system pre-calculates the DTW distance between a large number of unmatched signal pairs that have undergone Z-score normalization (i.e., dimensionless); since the physical units of the input signal have been removed, D dtw It becomes a purely numerical statistic; statistics show that for uncorrelated standardized random sequences, their cumulative distance grows linearly with time, with an expected value of 0.5 per second, or per unit normalized time step, which is dimensionless; therefore, the physical meaning of the coefficient 0.5 is the normalized background noise distance drift rate per unit time. Therefore, let T be set. obs ×0.5 is used as the half-decay point, meaning that when the distance between the real-time signal and the model reaches a level of random non-correlation, the similarity ρ... dtw Attenuation to This effectively distinguishes between similar and random states, thereby ensuring the accuracy of the calculated similarity value ρ. dtw It can stably distribute within the [0,1] interval, avoiding uncontrollable algorithm results due to undefined parameters.

[0027] Example 7: The process of determining the dominant type of signal deviation based on topological similarity and generating corresponding control instructions based on the dominant type includes: comparing the topological similarity value with a preset decision threshold; if the topological similarity value is greater than the decision threshold, determining that the current signal deviation is mainly dominated by theoretical interference characteristics; in response to this determination result, generating a first type of control instruction; the first type of control instruction is used to drive the actuator to start rhythmic guidance or quickly adjust the first environmental parameter.

[0028] This embodiment details the control strategy after determining that the insomnia is psychogenic; the system compares the calculated topological similarity value ρ with the preset decision threshold θ. th The comparison is performed; this threshold is typically set to 0.75. The specific method for obtaining this value is as follows: based on clinical labeled data samples containing 500 groups of anxiety-related and non-anxiety-related insomnia collected during the system development phase, receiver operating characteristic (ROC) curves are calculated under different similarity thresholds. The Youden index (sensitivity + specificity - 1) is selected, and the similarity value corresponding to the maximum point is used as the factory default threshold θ. th To ensure a high level of confidence in the judgment; if ρ>θ th If the current signal deviation is determined to be mainly due to theoretical interference characteristics, i.e., anxiety factors, then in response to this determination, the system generates a first type of control instruction, which aims to break the conditioned reflex of being awake while in bed. Specifically, the first type of control command is used to drive the actuator to initiate rhythmic guidance, such as driving the airbags in the mattress to produce a slight fluctuation of 0.1Hz to guide the breathing rate, or quickly adjusting the first environmental parameter, such as driving the temperature control module to reduce the temperature by 1.5 degrees Celsius within 2 minutes, using the biological principle of inducing sleepiness by lowering the core body temperature, and forcibly interrupting rumination of thought. In this embodiment, CBT-I therapy is automated in a sleep intervention and control scenario. When psychological anxiety is confirmed, the system adopts an active intervention strategy of guidance and blocking, rather than simple comfort adjustment. This strategy can effectively help users get out of a state of high arousal, demonstrating the concrete implementation of digital therapy at the hardware level and solving the problem that traditional devices cannot provide effective intervention when users are anxious.

[0029] Example 8: The process of determining the dominant type of signal deviation based on topological similarity and generating corresponding control commands based on the dominant type also includes: if the topological similarity value is less than the decision threshold, it is determined that the current signal deviation is mainly dominated by environmental noise or unsuitable physical conditions; in response to this determination result, a second type of control command is generated; the second type of control command is used to drive the actuator to adjust the physical support parameters or slowly adjust the second environmental parameters.

[0030] This embodiment details the control strategies after determining that the insomnia is not psychological; if the comparison results show that the topological similarity value ρ < θth If this indicates that although the user has a sleep deviation, it does not have a topological similarity to the anxiety model. Based on this, the system determines that the current signal deviation is mainly dominated by environmental noise or physical discomfort. In response to this determination, the system generates a second type of control command, which aims to eliminate physical interference. Specifically, the second type of control command is used to drive the actuator to adjust the physical support parameters, such as changing the support stiffness of the airbag assembly or redistributing pressure points to relieve local pressure, or slowly adjusting the second environmental parameters, such as fine-tuning the temperature at an extremely low rate of 0.5 degrees Celsius every 10 minutes to find the user's thermal comfort zone and avoid being awakened by sudden temperature changes. In this embodiment, a complete closed-loop control logic is formed in the environmental feedback adjustment scenario. When it is confirmed that the cause is not psychological, the system adopts an adaptive and compensatory strategy. By avoiding the erroneous activation of vibration guidance when the user is physically uncomfortable, the system minimizes the erroneous disturbance to the user and ensures the appropriateness and comfort of environmental control under different sleep disturbance sources.

[0031] 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 it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A sleep environment feedback control data processing method based on cognitive behavioral therapy logic, characterized in that, include: Collect non-contact multimodal time-series signals within the target monitoring area and obtain preset system reference parameters; Based on the system reference parameters, the reference signal balance curve over time is reconstructed, and a target signal manifold model characterizing the target state is constructed. The preset interference characteristics are converted into signal modulation operators, and the signal modulation operators are injected into the target signal manifold model to generate a simulated signal flow containing theoretical interference characteristics; A dual-track differential operation is performed on the non-contact multimodal time-series signal, the target signal manifold model, and the simulated signal stream to extract the first residual feature characterizing the current system signal deviation and the second residual feature characterizing the purely theoretical interference signal, respectively. Calculate the topological similarity between the first residual feature and the second residual feature in the time-frequency domain feature space; Based on the topological similarity, the dominant type of signal deviation is determined, and a corresponding first-type control command or second-type control command is generated according to the dominant type to drive the actuator to act, forming an environmental control closed loop.

2. The sleep environment feedback control data processing method based on cognitive behavioral therapy logic as described in claim 1, characterized in that, The non-contact multimodal time-series signals collected within the target monitoring area include: Vibration signals and periodic waveform data of the target area are collected using piezoelectric ceramic sensors or fiber optic sensor arrays. Dynamic micro-variation signals in the target area are acquired by using a pressure distribution matrix sampled at high frequency. The vibration signal, periodic waveform data, and dynamic micro-variable signal are cleaned and synchronized to generate a basic input data stream for subsequent processing.

3. The sleep environment feedback control data processing method based on cognitive behavioral therapy logic as described in claim 2, characterized in that, Constructing a target signal manifold model representing the target state includes: Within the target time window, the frequency of the dynamic micro-variable signal under the baseline state exhibits an exponential decay trend. The high-frequency components of the signal rhythm variability under the set reference state exhibit a preset rising slope. By combining the exponential decay trend and the rising slope, a reference signal model is generated, which serves as the target signal manifold model.

4. The sleep environment feedback control data processing method based on cognitive behavioral therapy logic as described in claim 3, characterized in that, The process involves converting preset interference characteristics into signal modulation operators and injecting these operators into the target signal manifold model to generate a simulated signal flow containing theoretical interference characteristics. Construct impulse noise that conforms to non-periodic abrupt changes as a first-class feature injection operator; A filter simulating a specific low-frequency suppression effect is constructed as a second-type feature injection operator; The first type of feature injection operator and the second type of feature injection operator are superimposed on the target signal manifold model to generate a simulated signal flow containing theoretical interference features.

5. The sleep environment feedback control data processing method based on cognitive behavioral therapy logic as described in claim 4, characterized in that, Perform dual-track differential operations on the non-contact multimodal time-series signal, the target signal manifold model, and the simulated signal stream, and extract the first residual feature and the second residual feature, respectively: The first difference between the basic input data stream and the target signal manifold model is calculated to obtain the first residual feature; Calculate the second difference between the simulated signal flow and the target signal manifold model to obtain the second residual characteristic; The first residual feature includes mixed information of environmental noise, physical state changes, and theoretical interference, while the second residual feature only includes signal features caused by the theoretical interference.

6. The sleep environment feedback control data processing method based on cognitive behavioral therapy logic as described in claim 5, characterized in that, Calculating the topological similarity between the first residual feature and the second residual feature in the time-frequency domain feature space includes: The first residual feature and the second residual feature are respectively subjected to time-frequency domain transformation, and the transformed feature coefficients are extracted. The distance between the feature coefficient sequences is calculated using a dynamic regularization algorithm, or the angle between the feature coefficients is calculated using a vector similarity algorithm, to obtain quantified topological similarity values.

7. The sleep environment feedback control data processing method based on cognitive behavioral therapy logic as described in claim 6, characterized in that, The dominant type of signal deviation is determined based on the topological similarity, and corresponding control instructions are generated according to the dominant type, including: The topological similarity value is compared with a preset decision threshold; If the topological similarity value is greater than the decision threshold, it is determined that the current signal deviation is mainly dominated by the theoretical interference features; In response to this determination, a first type of control instruction is generated; the first type of control instruction is used to drive the actuator to initiate rhythmic guidance or rapidly adjust the first environmental parameters.

8. The sleep environment feedback control data processing method based on cognitive behavioral therapy logic as described in claim 7, characterized in that, The process of determining the dominant type of signal deviation based on the topological similarity and generating corresponding control commands based on the dominant type further includes: If the topological similarity value is less than the decision threshold, it is determined that the current signal deviation is mainly dominated by environmental noise or unsuitable physical conditions. In response to this determination, a second type of control command is generated; the second type of control command is used to drive the actuator to adjust the physical support parameters or slowly adjust the second environmental parameters.

Images