Method and apparatus for multi-modal sensory self-adaptive regulation based on physiological characteristic feedback

Abstract

The application relates to a multi-modal sensory self-adaptive adjustment method and device based on physiological characteristic feedback, a computer device and a storage medium. The method comprises the following steps: determining a target physiological index to be adjusted, defining a sensory parameter space composed of multi-modal sensory control parameters and a control parameter vector; generating an initial control parameter vector based on a preset sampling strategy and converting the initial control parameter vector into a physical control instruction for execution, collecting corresponding physiological signals to generate a physiological characteristic vector; constructing a training set by combining the control parameter vector and the physiological characteristic vector into a data pair, establishing a probability agent model; performing parameter optimization iteration based on the model and the training set, selecting a target characteristic vector from a non-dominant physiological characteristic vector set and executing a physical control instruction corresponding to the target characteristic vector, and calculating an average physiological characteristic vector in a sliding time window to perform real-time monitoring and dynamically adjusting the physical control instruction. The application can realize personalized real-time adjustment of multi-modal environmental stimuli such as vision, hearing and olfaction, and effectively improve the user state.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and physiological signal processing technology, specifically to a multimodal sensory adaptive regulation method, device, computer equipment, and storage medium based on physiological feature feedback. Background Technology

[0002] In contemporary human-computer interaction and environmental regulation technologies, multimodal sensory stimulation is often used to regulate the user's environment in order to optimize the user's physiological state and sensory experience. Multimodal sensory regulation technology applies corresponding physical stimuli to users through multiple sensory channels such as vision, hearing, and smell to achieve specific physiological and psychological state regulation goals such as relaxation, focus, and sleep.

[0003] Traditional techniques for multimodal sensory modulation mainly fall into two categories: static parameter modulation, which provides constant sensory stimulation to users based on a preset combination of fixed parameters; and simple feedback modulation, which makes limited adjustments to a single sensory parameter based on a single physiological measurement of the user. Both of these traditional techniques rely on preset parameters or simple feedback as their core logic, and thus have significant limitations. Static parameter modulation cannot adapt to individual differences and dynamic changes in physiological states, resulting in inconsistent and unsustainable modulation effects. Simple feedback modulation is typically limited to a single sensory modality or a single physiological indicator, lacking a comprehensive balance across multiple physiological indicators, and is prone to modulation bias or overmodulation.

[0004] In existing technologies, multimodal sensory modulation systems typically employ open-loop control or simple closed-loop control mechanisms. Open-loop control systems rely entirely on preset parameters, failing to acquire real-time physiological feedback from the user, resulting in poor modulation accuracy and adaptability. While simple closed-loop control systems can acquire physiological feedback, they often employ threshold judgment and linear adjustment strategies, lacking effective modeling and optimization capabilities for complex nonlinear physiological response relationships, making it difficult to achieve optimal balance among multiple physiological indicators.

[0005] In multi-objective optimization, traditional techniques often employ linear methods such as weighted summation or priority ranking, which cannot effectively handle conflicts and trade-offs between objectives. For example, in relaxation state regulation, increasing EEG alpha wave power and improving heart rate variability may present parameter conflicts, and traditional methods struggle to find the optimal parameter combination that balances the two objectives.

[0006] In physiological signal processing, existing technologies mostly employ instantaneous sampling or fixed time windows for feature extraction, which cannot effectively eliminate the user's transient adaptive response and measurement noise. After being stimulated by senses, users typically require a certain adaptation period to reach a stable response. Traditional methods do not take this physiological characteristic into account, leading to inaccurate feature extraction and affecting the accuracy of regulatory decisions.

[0007] In summary, existing technologies suffer from the following technical problems: first, they lack effective adaptive optimization mechanisms for multimodal sensory parameter modulation; second, they lack the ability to balance and optimize multiple target physiological indicators; and third, physiological signal processing does not consider user adaptation characteristics, leading to inaccurate feature extraction. These problems limit the modulation accuracy, stability, and user experience of multimodal sensory modulation techniques. Therefore, a new multimodal sensory modulation method is needed to achieve adaptive optimization of multimodal sensory parameters by capturing the user's target physiological indicator values ​​in real time, thereby improving modulation accuracy and user experience. Summary of the Invention

[0008] To address the technical problems of poor adaptability, difficulty in balancing multiple objectives, and inaccurate physiological signal processing in traditional multimodal sensory regulation methods, this invention provides a multimodal sensory adaptive regulation method, device, computer equipment, and storage medium based on physiological feature feedback.

[0009] In a first aspect, the present invention provides a multimodal sensory adaptive regulation method based on physiological feature feedback, the method comprising the following steps:

[0010] The target physiological index to be adjusted is determined, and a sensory parameter space is defined. The sensory parameter space consists of the value ranges of M preset physical control parameters. The value range of each physical control parameter is mapped to a preset normalized numerical range.

[0011] Define a control parameter vector, which consists of M preset physical control parameters. Each component of the control parameter vector corresponds to a physical control parameter, and the physical control parameters include multiple sensory modal control parameters.

[0012] Within the sensory parameter space, N preset control parameter vectors are generated based on a preset sampling strategy. For each control parameter vector, the control parameter vector is converted into a physical control instruction through a preset hardware mapping logic and executed. After the physical control instruction corresponding to the control parameter vector is executed, the corresponding physiological signal is collected and the corresponding physiological feature vector is generated. Each component of the physiological feature vector corresponds to a target physiological indicator.

[0013] Each control parameter vector and its corresponding physiological feature vector are combined into an input-output data pair, which forms a training set. A probabilistic proxy model is then constructed based on the training set.

[0014] Based on the probabilistic proxy model, the training set, and the sensory parameter space, a parameter optimization iterative process is executed until the iteration stopping condition is met.

[0015] When the iteration stopping condition is met, a set of non-dominated physiological feature vectors is selected from the training set. The dominance relationship is defined as follows: for any two physiological feature vectors, if every target physiological index of the first vector is not inferior to that of the second vector, and at least one index is superior to that of the second vector, then the first vector is said to dominate the second vector. The set of non-dominated physiological feature vectors in the training set refers to the set of physiological feature vectors that are not dominated by other physiological feature vectors in the training set.

[0016] A physiological feature vector is selected from the set of non-dominated physiological feature vectors using a preset physiological feature vector selection strategy. The control parameter vector corresponding to the selected physiological feature vector in the training set is converted into a physical control command through a preset hardware mapping logic and executed. During execution, time windows are divided by a preset step size. The arithmetic mean of the physiological feature vectors in each sliding time window is calculated as the average physiological feature vector of the sliding time window. The current physical control command is adjusted according to the training set, the set of non-dominated physiological feature vectors, and the average physiological feature vector of the most recent sliding time window.

[0017] In one embodiment, the step of converting the control parameter vector into physical control instructions and executing them through a preset hardware mapping logic includes: processing each component of the control parameter vector one by one; if the component is a visual control parameter, converting the component into a visual control instruction; if the component is an auditory control parameter, converting the component into an auditory control instruction; and if the component is an olfactory control parameter, converting the component into an olfactory control instruction.

[0018] In one embodiment, the step of acquiring corresponding physiological signals and generating corresponding physiological feature vectors after the execution of the physical control command corresponding to the control parameter vector includes: continuously receiving physiological signals but not including them in feature calculation during a first preset duration adaptation buffer period after the execution of the physical control command; after the adaptation buffer period ends, opening a feature extraction sliding window of a second preset duration and updating the feature extraction sliding window by sliding with a preset step size; performing bandpass filtering on the original waveform data in the feature extraction sliding window on a preset feature frequency band, and calculating the arithmetic mean of each feature data after filtering to form a physiological feature vector, wherein each component of the physiological feature vector is the arithmetic mean of a feature data.

[0019] In one embodiment, the step of performing parameter optimization iterative processes based on the probabilistic proxy model, training set, and sensory parameter space until the iteration stopping condition is met includes:

[0020] The probabilistic surrogate model is a Gaussian process regression model;

[0021] Each iteration of the parameter optimization iterative process consists of four steps: Step 1: Select a set of non-dominated physiological feature vectors from the training set; Step 2: Based on the probabilistic surrogate model and the set of non-dominated physiological feature vectors, determine the control parameter vector with the largest expected hypervolume increment in the sensory parameter space. The expected hypervolume increment corresponding to the control parameter vector refers to the difference between the measure of the area covered by the updated set of non-dominated physiological feature vectors relative to the preset reference point and the measure of the area covered by the original set of non-dominated physiological feature vectors relative to the preset reference point. The updated set of non-dominated physiological feature vectors refers to the new set of non-dominated physiological feature vectors obtained by applying the probabilistic surrogate model to the control parameter vectors, obtaining the predicted physiological feature vectors, and adding the predicted physiological feature vectors to the original set of non-dominated physiological feature vectors; Step 3: Convert the control parameter vector with the largest expected hypervolume increment into a physical control command through a preset hardware mapping logic and execute it to collect the corresponding physiological signals and generate the corresponding physiological feature vector; Step 4: Form an input-output data pair with the control parameter vector with the largest expected hypervolume increment and its corresponding physiological feature vector, add the input-output data pair to the training set, and reconstruct the probabilistic surrogate model based on the training set;

[0022] The iteration stopping condition is when the number of iterations reaches a preset maximum value or the maximum expected supervolume increment is lower than a preset threshold.

[0023] In one embodiment, adjusting the current physical control command based on the training set, the set of non-dominated physiological feature vectors, and the average physiological feature vector of the most recent sliding time window includes: if any component of the average physiological feature vector deviates from the preset target value of the corresponding target physiological indicator by more than a preset severe deterioration threshold, then the parameter optimization iteration process is re-executed based on the probabilistic proxy model, the training set, and the sensory parameter space until the iteration stopping condition is met; when the iteration stopping condition is met, the set of non-dominated physiological feature vectors is selected from the training set, and a physiological feature vector is selected using a preset physiological feature vector selection strategy, and the selected physiological feature vector is then used in the training set. The control parameter vector corresponding to the training set is converted into a physical control command through a preset hardware mapping logic, which serves as the adjusted current physical control command. If the average physiological feature vector is dominated by any vector in the set of non-dominated physiological feature vectors, a physiological feature vector that dominates the average physiological feature vector is selected from the set of non-dominated physiological feature vectors using a preset physiological feature vector selection strategy. This is denoted as the target physiological feature vector, and the control parameter vector corresponding to the target physiological feature vector in the training set is converted into a physical control command through a preset hardware mapping logic, which serves as the adjusted current physical control command. If neither of the above two conditions is met, the current physical control command remains unchanged.

[0024] Secondly, this invention provides a multimodal sensory adaptive adjustment device based on physiological feature feedback. The device includes: an initialization module, used to determine the target physiological index to be adjusted and define a sensory parameter space and control parameter vectors. Within the sensory parameter space, N preset control parameter vectors are generated based on a preset sampling strategy. For each control parameter vector, the control parameter vector is converted into a physical control instruction through a preset hardware mapping logic and executed. After the physical control instruction corresponding to the control parameter vector is executed, the corresponding physiological signal is collected and a corresponding physiological feature vector is generated. Finally, each control parameter vector and its corresponding physiological feature vector are combined into an input-output data pair to form a training set. An optimization iteration module is used to construct a probabilistic proxy model based on the training set and, based on the probability... The surrogate model, training set, and sensory parameter space execute parameter optimization iterative processes until the iteration stop condition is met. The real-time monitoring and adjustment module is used to select a set of non-dominated physiological feature vectors from the training set, select a physiological feature vector from the set of non-dominated physiological feature vectors using a preset physiological feature vector selection strategy, convert the control parameter vector corresponding to the selected physiological feature vector in the training set into a physical control command through a preset hardware mapping logic and execute it. During execution, the time window is divided by a preset step size, and the arithmetic mean of the physiological feature vectors in each sliding time window is calculated as the average physiological feature vector of the sliding time window. The current physical control command is adjusted according to the training set, the set of non-dominated physiological feature vectors, and the average physiological feature vector of the most recent sliding time window.

[0025] Thirdly, the present invention provides a computer device, including a memory, a processor, a physiological signal acquisition module, a sensory stimulation generation module, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described in any one of claims 1 to 5.

[0026] In one embodiment, the physiological signal acquisition module is used to acquire the user's raw physiological signals in real time; the physiological signal acquisition module includes at least one of an electroencephalogram (EEG) acquisition unit, an electrocardiogram (ECG) acquisition unit, a skin conductance acquisition unit, and a blood oxygen acquisition unit.

[0027] In one embodiment, the sensory stimulation generation module is used to execute physical control commands to generate environmental stimuli; the sensory stimulation generation module includes at least one of a visual stimulation unit, an auditory stimulation unit, and an olfactory stimulation unit.

[0028] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method according to any one of claims 1 to 5.

[0029] The aforementioned multimodal sensory adaptive regulation method, device, computer equipment, and readable storage medium based on physiological feature feedback have the following main technical effects. First, by constructing a probabilistic surrogate model and an expected hypervolume incremental optimization algorithm, this invention can achieve adaptive balance optimization among multiple physiological indicators, overcoming the limitations of traditional methods in multi-objective optimization and effectively handling conflicts and trade-offs between objectives. Second, by setting an adaptation buffer period and a sliding time window for physiological signal feature extraction, it effectively eliminates the user's instantaneous adaptive response and measurement noise, improving the accuracy of feature extraction and the stability of regulation, and avoiding regulation deviations caused by instantaneous responses. Third, through real-time monitoring and dynamic regulation mechanisms, regulation deviations can be detected and corrected in a timely manner. When a physiological state is detected to deviate significantly from the target or be dominated by a non-dominated set, automatic re-optimization or selection of a new control parameter vector is triggered, ensuring the continuity of the regulation effect and user satisfaction. Finally, through independent mapping and physical control of multimodal sensory parameters, a personalized sensory regulation experience is achieved, meeting the physiological and psychological needs of different users and improving the accuracy of regulation and user experience. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a flowchart illustrating a multimodal sensory adaptive regulation method based on physiological feature feedback in one embodiment.

[0032] Figure 2 This is a schematic diagram of the parameter optimization iteration process in a multimodal sensory adaptive regulation method based on physiological feature feedback in one embodiment;

[0033] Figure 3 This is a structural block diagram of a multimodal sensory adaptive adjustment device based on physiological feature feedback in one embodiment;

[0034] Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0036] It should be noted that the terms "first," "second," etc., used in this invention can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this invention, are intended to cover non-exclusive inclusion. The term "a plurality of" used in this invention refers to two or more. The term "and / or" used in this invention refers to one of the embodiments, or any combination of multiple embodiments.

[0037] like Figure 1 As shown, this invention provides a multimodal sensory adaptive regulation method based on physiological feature feedback, which includes the following steps:

[0038] Step S102: Determine the target physiological indicator to be adjusted and define the sensory parameter space. The sensory parameter space consists of the value ranges of M preset physical control parameters. The value range of each physical control parameter is mapped to a preset normalized numerical range.

[0039] Among them, the target physiological indicators are physiological state parameters that the user wishes to adjust, including but not limited to at least one of the following: brainwave alpha power, brainwave beta power, brainwave theta power, brainwave delta power, heart rate variability (HRV), skin conductance response (GSR), blood oxygen saturation (SpO2), respiratory rate, and body temperature.

[0040] The sensory parameter space consists of the value ranges of M physical control parameters, where M is an integer greater than or equal to 1. The value range of each physical control parameter is pre-mapped to a normalized numerical interval [0, 1], where 0 corresponds to the minimum value and 1 corresponds to the maximum value. The physical control parameters include visual, auditory, and olfactory control parameters. Visual control parameters include LED brightness parameters (range 0-100%) and LED color temperature parameters (range 2700K-6500K); auditory control parameters include volume gain parameters (range -20dB to +20dB) and audio frequency parameters (range 100Hz-1000Hz); olfactory control parameters include fragrance concentration parameters (range 0-100%) and atomization intensity parameters (range 0-100%). Normalizing different physical control parameters to a unified interval facilitates subsequent optimization calculations and parameter adjustments.

[0041] In one specific embodiment, taking the adjustment of a user's relaxation state as an example, the target physiological indicators to be adjusted are EEG alpha wave power and heart rate variability. EEG alpha wave power reflects the degree of brain relaxation, with a normal range of 8-13 μV. 2Heart rate variability reflects the balance of the autonomic nervous system, with a normal range of 40-60 ms. The sensory parameter space consists of the value ranges of M = 6 physical control parameters, each mapped to the interval [0, 1]. The control parameter vector is x = [x1, x2, x3, x4, x5, x6], where x1 corresponds to the LED brightness parameter, x2 to the LED color temperature parameter, x3 to the volume gain parameter, x4 to the audio frequency parameter, x5 to the fragrance concentration parameter, and x6 to the atomization intensity parameter.

[0042] Step S104: Define a control parameter vector. The control parameter vector consists of M preset physical control parameters. Each component of the control parameter vector corresponds to a physical control parameter, which includes visual control parameters, auditory control parameters, and olfactory control parameters.

[0043] In this context, the control parameter vector is a point in the sensory parameter space, representing a specific combination of physical control parameters. Each component of the control parameter vector corresponds to a physical control parameter, and the value of each component lies within the interval [0, 1]. By defining a multidimensional control parameter vector, precise control of multimodal sensory stimuli can be achieved.

[0044] In one specific embodiment, the control parameter vector is a 6-dimensional vector x = [x1, x2, x3, x4, x5, x6], where the first two components correspond to visual control parameters, the third and fourth components correspond to auditory control parameters, and the fifth and sixth components correspond to olfactory control parameters.

[0045] Step S106: Within the sensory parameter space, N preset control parameter vectors are generated based on a preset sampling strategy. For each control parameter vector, the control parameter vector is converted into a physical control instruction through a preset hardware mapping logic and executed. After the physical control instruction corresponding to the control parameter vector is executed, the corresponding physiological signal is collected and a corresponding physiological feature vector is generated. Each component of the physiological feature vector corresponds to a target physiological indicator.

[0046] The preset sampling strategy can be Latin hypercube sampling, uniform sampling, or random sampling, etc., to ensure uniform coverage in the sensory parameter space.

[0047] In one specific embodiment, the preset sampling strategy employs Latin hypercube sampling. Latin hypercube sampling can achieve uniform coverage in a high-dimensional parameter space, ensuring the diversity of the initial samples.

[0048] The default hardware mapping logic converts each component of the control parameter vector into a corresponding physical control command.

[0049] In one specific embodiment, visual control parameters are mapped to the PWM dimming duty cycle of the LED driver circuit or the LED color temperature control signal. Specifically, the LED brightness parameter x1 is linearly mapped to the duty cycle of the PWM signal, with a duty cycle range of 0%-100%, corresponding to brightness from dark to brightest. The LED color temperature parameter x2 is mapped to the LED color temperature control signal, and the color temperature is adjusted from 2700K to 6500K by adjusting the current ratio of the warm and cool LEDs. Auditory control parameters are mapped to the decibel gain or frequency signal of the audio system. Specifically, the volume gain parameter x3 is mapped to the gain value of the audio amplifier, with a gain range of -20dB to +20dB. The audio frequency parameter x4 is mapped to the frequency value of the audio generator, with a frequency range of 100Hz-1000Hz. Olfactory control parameters are mapped to the opening duration of the pump valve or the atomization duty cycle in the fragrance generator. Specifically, the fragrance concentration parameter x5 is mapped to the opening duration of the fragrance pump valve, ranging from 0 to 5 seconds. The longer the opening duration, the greater the amount of fragrance released. The atomization intensity parameter x6 is mapped to the duty cycle of the ultrasonic atomizer, ranging from 0% to 100%. The higher the duty cycle, the greater the atomization intensity.

[0050] The physiological signal acquisition process includes: after the physical control command is executed, physiological signals are continuously received during the first preset duration of the adaptation buffer period but are not included in the feature calculation; after the adaptation buffer period ends, a feature extraction sliding window of the second preset duration is opened and the feature extraction sliding window is updated by sliding with a preset step size; the original waveform data in the feature extraction sliding window is subjected to bandpass filtering of the preset feature frequency band, and the arithmetic mean of each feature data after filtering is calculated to form the physiological feature vector.

[0051] The purpose of the adaptation buffer period is to provide users with an adaptation time and avoid noise caused by transient responses. After the physical control command is executed, the user needs a certain amount of time to adapt to the new sensory environment. During this period, the physiological signals collected may not accurately reflect the user's stable physiological state. Therefore, physiological signals are continuously received during the adaptation buffer period but not included in feature calculation to ensure the accuracy of subsequent feature extraction.

[0052] The feature extraction sliding window is used to collect physiological signals after the user has stabilized and adapted. After the adaptation buffer period ends, the feature extraction sliding window is opened for a second preset duration and updated by sliding with a preset step size to capture dynamic changes in physiological signals.

[0053] Bandpass filtering is used to extract physiological signal features in specific frequency bands. For electroencephalogram (EEG) signals, an 8-13Hz bandpass filter is used to extract alpha waves, reflecting the degree of brain relaxation. For electrocardiogram (ECG) signals, a 0.05-100Hz bandpass filter is used to remove 50Hz power frequency interference, and the standard deviation of the RR interval is calculated as an indicator of heart rate variability, reflecting the balance state of the autonomic nervous system. For electrodermal conductance (EDC) signals, a 0.05-5Hz bandpass filter is used, and the arithmetic mean of skin conductance levels is calculated, reflecting the level of psychophysiological arousal. For blood oxygenation signals, a moving average filter is used, and the arithmetic mean of blood oxygen saturation is calculated, reflecting the blood oxygenation status.

[0054] In one specific embodiment, Latin hypercube sampling is used to generate N=20 uniformly distributed sampling points in a 6-dimensional sensory parameter space. For each sampling point, hardware mapping logic is used to convert it into physical control instructions and execute them. For example, the physical control instructions corresponding to the first sampling point x1=[0.2, 0.3, 0.5, 0.7, 0.4, 0.6] are: LED brightness 20%, LED color temperature 30% (corresponding to 3500K), volume gain 0dB, audio frequency 700Hz, fragrance concentration 40%, and atomization intensity 60%. After each physical control command is executed, a first preset duration of 30 seconds for adaptation buffering is set. During this period, physiological signals are continuously received but not included in feature calculation to eliminate the user's instantaneous adaptation response. After the adaptation buffering period ends, a second preset duration of 60 seconds for feature extraction sliding window is opened and updated by sliding the feature extraction sliding window in preset steps of 10 seconds. The raw EEG signals in each sliding window are bandpass filtered at 8-13Hz to extract alpha waves, and the arithmetic mean of the alpha wave power is calculated. The ECG signals in each sliding window are subjected to R wave detection, and the standard deviation of the RR interval is calculated as an index of heart rate variability. The alpha wave power and heart rate variability calculated in each sliding window are combined to form a physiological feature vector y = [α, HRV].

[0055] Step S108: Each control parameter vector and its corresponding physiological feature vector are combined into an input-output data pair, the input-output data pairs form a training set, and a probabilistic surrogate model is constructed based on the training set.

[0056] In one specific embodiment, the training set is denoted as D = {(xi, yi)}, i = 1, 2, ..., N, where x i Let y be the vector of the i-th control parameter. iThe corresponding physiological feature vector is represented by y. The probabilistic surrogate model employs Gaussian Process Regression (GPR) and optimizes the hyperparameters using the maximum likelihood estimation method to obtain a probabilistic surrogate model capable of predicting the mapping relationship between the physiological feature vector and the control parameter vector. For any control parameter vector x, the GPR model outputs the mean μ(x) and variance ∑(x) of the predicted physiological feature vector, i.e., y ~ N(μ(x), ∑(x)), thereby quantifying the uncertainty of the prediction.

[0057] In one specific embodiment, 20 control parameter vectors and their corresponding physiological feature vectors are combined into input-output data pairs to form a training set. A Gaussian process regression model is constructed based on the training set, using the radial basis function (RBF) as the kernel function. The kernel function expression is as follows: Where σ 2 Let L be the signal variance and L be the length scale vector. The hyperparameters are optimized using the maximum likelihood estimation method to obtain optimal hyperparameter values, such as σ. 2 =1.2, L = [0.8, 0.6, 0.9, 0.7, 0.5, 0.8]. A well-trained GPR model can predict the mean and variance of the physiological feature vectors corresponding to any control parameter vector.

[0058] Step S110: Based on the probabilistic proxy model, the training set, and the sensory parameter space, perform a parameter optimization iterative process until the iteration stopping condition is met.

[0059] like Figure 2 As shown, each iteration in the parameter optimization iterative process consists of four steps:

[0060] Step 1: Select the set of non-dominated physiological feature vectors from the training set. The Non-Dominated Sorting Algorithm (NSGA-II) is used to sort the physiological feature vectors in the training set to obtain the non-dominated front. The set of non-dominated physiological feature vectors refers to the set of physiological feature vectors that are not dominated by other physiological feature vectors in the training set. The dominance relationship is defined as follows: for any two physiological feature vectors y1 and y2, if every target physiological indicator of y1 is no worse than y2, and at least one indicator is better than y2, then y1 is said to dominate y2.

[0061] Step two involves determining the control parameter vector that maximizes the Expected Hypervolume Improvement (EHVI) in the sensory parameter space based on a probabilistic surrogate model and a set of non-dominated physiological feature vectors. The EHVI is a collection function constructed based on a Gaussian process regression (GPR) model, which effectively handles probabilistic predictions for multi-objective optimization, achieving an intelligent balance between exploration and utilization. Specifically, a preset reference point is first defined as [r1, r2, ..., r...]. k [k], where k is the number of target physiological indicators. Then, for candidate control parameter vectors in the sensory parameter space, the mean and variance of their physiological feature vectors are predicted using the GPR model. Based on the predicted mean and variance, the expected hypervolume increment is calculated using the Monte Carlo method. Finally, the control parameter vector x that maximizes the expected hypervolume increment is selected. * .

[0062] Step 3 involves maximizing the control parameter vector x that represents the desired hypervolume increment. * The hardware mapping logic is converted into physical control instructions and executed, and the corresponding physiological signals are collected and the corresponding physiological feature vector y is generated. * .

[0063] Step four: (x) * y * The input-output data pairs are formed and added to the training set. The probabilistic proxy model is then reconstructed based on the updated training set.

[0064] The iteration stops when the number of iterations reaches a preset maximum value or the maximum expected supervolume increment is lower than a preset threshold.

[0065] Among them, Gaussian Process Regression (GPR) is a non-parametric probabilistic model capable of Bayesian inference of functions. The GPR model assumes that the function values ​​follow a Gaussian process and uses a kernel function to describe the correlation between function values. In this invention, a radial basis function (RBF) is used as the kernel function, and its expression is: Where σ 2 Let be the signal variance and L be the length scale vector. For any control parameter vector x, the GPR model outputs the mean μ(x) and variance ∑(x) of the predicted physiological feature vector, thereby quantifying the uncertainty of the prediction.

[0066] Expected Hypervolume Improvement (EHVI) is a multi-objective optimization acquisition function used to evaluate the expected hypervolume improvement after sampling candidate points. EHVI comprehensively considers the degree to which the predicted mean approaches the current non-dominated frontier (utilizing known preferred regions) and the magnitude of the predicted variance (exploring unknown regions), thereby achieving a balance between exploration and utilization.

[0067] The EHVI is approximated using the Monte Carlo method. For a candidate control parameter vector x, the GPR model outputs the mean μ(x) and variance ∑(x) of the predicted physiological feature vector. S = 1000 samples y are sampled from the predicted distribution. (s) ~N(μ(x), ∑(x)), s=1,2,...,1000. For each sample y (s) Determine whether it is dominated by any vector in the current non-dominated physiological feature vector set Y. If it is not dominated, then set y (s) Add to update set Y∪{y (s)}, and compute the update set Y∪{y (s) The hypervolume HV(Y∪{y} relative to the preset reference point r) (s)}), and the hypervolume HV(Y) of the original non-dominated set Y. The expected hypervolume increment is expressed by the formula EHVI(x)=(1 / S)·∑ s＝1 ..., s max(0, HV(Y∪{y (s) The calculation is performed using EHVI(x). The optimal control parameter vector x* is selected by maximizing EHVI(x).

[0068] In one specific embodiment, the iteration stopping condition is set to 50 iterations or the maximum expected hypervolume increment being less than 0.001. In each iteration, a set of non-dominated physiological feature vectors is first selected from the training set. Then, based on the GPR model and the non-dominated set, the control parameter vector with the maximum expected hypervolume increment is calculated. This control parameter vector is executed, and new physiological data is collected. The new data is added to the training set, and the GPR model is retrained. After 50 iterations, the training set contains 70 data points, the set of non-dominated physiological feature vectors contains 15 vectors, and the non-dominated front covers the alpha wave power range of 6.5-9.2 μV. 2 HRV range 45-58ms.

[0069] Step S112: After the iteration stopping condition is met, a set of non-dominated physiological feature vectors is selected from the training set. The dominance relationship is defined as follows: for any two physiological feature vectors, if every target physiological indicator of the first vector is not inferior to that of the second vector, and at least one indicator is superior to that of the second vector, then the first vector is said to dominate the second vector. The set of non-dominated physiological feature vectors in the training set refers to the set of physiological feature vectors that are not dominated by other physiological feature vectors in the training set.

[0070] In one specific embodiment, a fast non-dominated sorting algorithm is used to sort the physiological feature vectors in the final training set and filter out the non-dominated physiological feature vector set. The non-dominated physiological feature vector set refers to the set of physiological feature vectors that are not dominated by other physiological feature vectors in the training set, representing the physiological feature state that achieves the best balance among multiple target physiological indicators.

[0071] In one specific embodiment, after 50 iterations, the training set contains 70 input-output data pairs, from which 15 non-dominated physiological feature vectors are selected. These non-dominated vectors are not dominated by any other vectors in terms of both alpha wave power and heart rate variability.

[0072] Step S114: Select a physiological feature vector from the set of non-dominated physiological feature vectors using a preset physiological feature vector selection strategy; convert the control parameter vector corresponding to the selected physiological feature vector in the training set into a physical control command through a preset hardware mapping logic and execute it; during execution, divide the time window by sliding with a preset step size, calculate the arithmetic mean of the physiological feature vectors in each sliding time window as the average physiological feature vector of the sliding time window, and adjust the current physical control command according to the training set, the set of non-dominated physiological feature vectors and the average physiological feature vector of the most recent sliding time window.

[0073] The preset physiological feature vector selection strategy can be a lexicographical sorting strategy, a weighted scoring strategy, or a priority rule strategy. The lexicographical sorting strategy selects features according to the priority order of their components, prioritizing the optimal values ​​of the main target physiological indicators. The weighted scoring strategy performs a comprehensive scoring and sorting of the physiological feature vectors based on preset weights, calculating the physiological feature vector y = [y1, y2, ..., y...]. k The overall score S = ∑w i ·y i , where w i The weights of the i-th target physiological indicator are sorted in descending order of score, and the physiological feature vector with the highest score is selected. The priority rule strategy selects the optimal value based on the preset priority rules of the target physiological indicators, prioritizing the optimal value of high-priority targets.

[0074] During the execution of physical control commands, real-time monitoring and dynamic adjustment are performed, including: if any component of the average physiological feature vector deviates from the preset target value of the corresponding target physiological indicator by more than a preset severe deterioration threshold, then the parameter optimization iteration process is re-executed based on the probabilistic proxy model, the training set, and the sensory parameter space until the iteration stopping condition is met; when the iteration stopping condition is met, a set of non-dominated physiological feature vectors is selected from the training set, and a physiological feature vector is selected from the set of non-dominated physiological feature vectors using a preset physiological feature vector selection strategy. The control parameter vector corresponding to the selected physiological feature vector in the training set is converted into a physical control command through a preset hardware mapping logic, which is used as the adjusted current physical control command; if the average physiological feature vector is dominated by any vector in the set of non-dominated physiological feature vectors, then a physiological feature vector dominating the average physiological feature vector is selected from the set of non-dominated physiological feature vectors using a preset physiological feature vector selection strategy, denoted as the target physiological feature vector, and the control parameter vector corresponding to the target physiological feature vector in the training set is converted into a physical control command through a preset hardware mapping logic, which is used as the adjusted current physical control command; if neither of the above two conditions is met, then the current physical control command remains unchanged.

[0075] In one specific embodiment, a time window is divided by sliding with a preset step size Δt. Each time window has a length of T. Within each sliding time window, the user's physiological signals are continuously collected and physiological feature vectors are calculated. Then, the arithmetic mean of all physiological feature vectors within the sliding time window is calculated as the average physiological feature vector y_avg of the sliding time window.

[0076] The system monitors the following three scenarios in real time and dynamically adjusts the corresponding physical control commands based on each scenario.

[0077] Scenario 1: If any component of the average physiological feature vector y_avg deviates from the preset target value of the corresponding target physiological indicator by more than a preset severe deterioration threshold, the parameter optimization iteration process is re-triggered. For example, when the target value of the alpha wave power is 8μV... 2 When the severe degradation threshold is 30%, if the detected α-wave power component of y_avg is lower than 5.6μV... 2 or higher than 10.4 μV 2If the user's relaxed state deviates significantly from the target, the physical control parameters need to be re-optimized. Specifically, the parameter optimization iterative process is re-executed based on the current probabilistic proxy model, training set, and sensory parameter space until the iteration stopping condition is met. Once the iteration stopping condition is met, a set of non-dominated physiological feature vectors is selected from the training set, and a physiological feature vector is selected from it using a preset physiological feature vector selection strategy. The control parameter vector corresponding to the selected physiological feature vector in the training set is converted into a physical control command through a preset hardware mapping logic, which serves as the adjusted physical control command.

[0078] In the second scenario, if the average physiological feature vector y_avg is dominated by any vector in the set of non-dominated physiological feature vectors, then a physiological feature vector y_target that dominates y_avg is selected from the set of non-dominated physiological feature vectors using a preset physiological feature vector selection strategy, and the control parameter vector corresponding to y_target is converted into a physical control command as the adjusted physical control command.

[0079] Scenario 3: If neither of the above two conditions is met, then the physical control command remains unchanged.

[0080] In one specific embodiment, the preset physiological feature vector selection strategy adopts a weighted scoring strategy. The α-wave power weight w_α = 0.7 and the HRV weight w_HRV = 0.3 are set. The comprehensive score S = 0.7α + 0.3HRV for each non-dominated physiological feature vector y = [α, HRV] is calculated. The physiological feature vector with the highest score y_opt = [8.5μV] is selected. 2 The corresponding control parameter vector is x_opt = [0.4, 0.5, 0.55, 0.6, 0.45, 0.5]. x_opt is converted into physical control commands and executed: LED brightness 40%, LED color temperature 50% (corresponding to 4600K), volume gain +5.5dB, audio frequency 600Hz, fragrance concentration 45%, and atomization intensity 50%. During execution, a 60-second time window is divided with 10-second steps, and the arithmetic mean of the physiological characteristic vector within each window is calculated. Monitoring revealed that at the 15th minute, the average alpha wave power dropped to 6.2μV. 2 The value deviates from the target value by 8 μV. 2 The deviation reached 22.5%, exceeding the severe deterioration threshold of 30%, triggering a re-optimization process. The parameter optimization iteration was re-executed, and after 15 iterations, a new optimal control parameter vector x_new = [0.38, 0.48, 0.58, 0.62, 0.42, 0.48] was found. After executing the new control command, the alpha wave power recovered to 8.3 μV. 2 HRV remained at 51ms, and the system stabilized.

[0081] like Figure 3 As shown, the present invention also provides a multimodal sensory adaptive adjustment device based on physiological feature feedback, the device comprising:

[0082] An initialization module is used to determine the target physiological index to be adjusted and to define a sensory parameter space and a control parameter vector. Within the sensory parameter space, N preset control parameter vectors are generated based on a preset sampling strategy. For each control parameter vector, the control parameter vector is converted into a physical control instruction through a preset hardware mapping logic and executed. After the physical control instruction corresponding to the control parameter vector is executed, the corresponding physiological signal is collected and the corresponding physiological feature vector is generated. Finally, each control parameter vector and its corresponding physiological feature vector are combined into an input-output data pair to form a training set.

[0083] The optimization iteration module is used to construct a probabilistic proxy model based on the training set, and to execute a parameter optimization iteration process based on the probabilistic proxy model, the training set, and the sensory parameter space until the iteration stopping condition is met.

[0084] The real-time monitoring and adjustment module is used to select a set of non-dominated physiological feature vectors from the training set, select a physiological feature vector from the set of non-dominated physiological feature vectors using a preset physiological feature vector selection strategy, convert the control parameter vector corresponding to the selected physiological feature vector in the training set into a physical control command through a preset hardware mapping logic and execute it, divide the time window by a preset step size during execution, calculate the arithmetic mean of the physiological feature vectors in each sliding time window as the average physiological feature vector of the sliding time window, and adjust the current physical control command according to the training set, the set of non-dominated physiological feature vectors and the average physiological feature vector of the most recent sliding time window.

[0085] In one specific embodiment, the initialization module includes: a target physiological index determination submodule, used to determine the target physiological index to be regulated; a sensory parameter space definition submodule, used to define a sensory parameter space consisting of the value ranges of M physical control parameters; a control parameter vector definition submodule, used to define a control parameter vector consisting of M physical control parameters; a sampling submodule, used to generate N control parameter vectors in the sensory parameter space based on a preset sampling strategy; a control instruction conversion submodule, used to convert the control parameter vectors into physical control instructions through preset hardware mapping logic and execute them; a physiological signal acquisition submodule, used to acquire physiological signals after the execution of physical control instructions and generate physiological feature vectors; and a training set construction submodule, used to combine the control parameter vectors and physiological feature vectors into input-output data pairs to form a training set.

[0086] The optimization iteration module includes: a probabilistic surrogate model construction submodule, used to construct a Gaussian process regression model based on the training set; a non-dominated set selection submodule, used to select a set of non-dominated physiological feature vectors from the training set; a desired hypervolume increment calculation submodule, used to determine the control parameter vector that maximizes the desired hypervolume increment in the sensory parameter space based on the probabilistic surrogate model and the non-dominated physiological feature vector set; an iteration execution submodule, used to execute the parameter optimization iteration process, including executing the control parameter vector, collecting physiological signals, updating the training set, and reconstructing the probabilistic surrogate model; and a stopping condition judgment submodule, used to determine whether the iteration stopping condition is met.

[0087] The real-time monitoring and adjustment module includes: a non-dominated set filtering submodule, used to filter out a set of non-dominated physiological feature vectors from the training set; a physiological feature vector selection submodule, used to select a physiological feature vector from the set of non-dominated physiological feature vectors using a preset physiological feature vector selection strategy; a control command execution submodule, used to convert the control parameter vectors corresponding to the selected physiological feature vectors into physical control commands and execute them; a sliding time window calculation submodule, used to divide the time window by a preset step size and calculate the arithmetic mean of the physiological feature vectors within each sliding time window; and an adjustment rule judgment submodule, used to adjust the current physical control command according to the training set, the set of non-dominated physiological feature vectors, and the average physiological feature vector, according to a preset adjustment rule.

[0088] like Figure 4 As shown, the present invention also provides a computer device, including a memory, a processor, a physiological signal acquisition module, a sensory stimulation generation module, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the above-described method embodiments.

[0089] The physiological signal acquisition module is used to acquire the user's raw physiological signals in real time, including at least one of the following: electroencephalogram (EEG) acquisition unit, electrocardiogram (ECG) acquisition unit, skin conductance acquisition unit, and blood oxygenation acquisition unit. The sensory stimulation generation module is used to execute physical control commands to generate environmental stimuli, including at least one of the following: visual stimulation unit, auditory stimulation unit, and olfactory stimulation unit.

[0090] In one specific embodiment, the EEG acquisition unit employs an 8-channel EEG amplifier with a sampling frequency of 250Hz and a bandpass filter range of 0.5-40Hz to acquire the user's EEG signals and extract features such as alpha, beta, theta, and delta waves. The ECG acquisition unit employs a single-lead ECG amplifier with a sampling frequency of 500Hz and a bandpass filter range of 0.05-100Hz to acquire the user's ECG signals and calculate indicators such as heart rate variability. The skin conductance acquisition unit employs a skin conductance sensor with a sampling frequency of 50Hz and a bandpass filter range of 0.05-5Hz to acquire the user's skin conductance signals and extract features such as skin conductance response levels. The blood oxygen acquisition unit employs a pulse oximeter with a sampling frequency of 100Hz to acquire the user's blood oxygen saturation signal.

[0091] In one specific embodiment, the visual stimulation unit uses an RGBW LED strip, supporting PWM dimming and color temperature adjustment, and is connected to the main control unit via a GPIO interface. The auditory stimulation unit uses a high-fidelity audio player, supporting volume, frequency, and binaural beat adjustment, and is connected to the main control unit via an I2S interface. The olfactory stimulation unit uses a multi-channel fragrance generator, supporting concentration and atomization intensity adjustment, and is connected to the main control unit via an I2C interface.

[0092] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps in the above-described method embodiments.

[0093] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0094] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0095] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A multimodal sensory adaptive regulation method based on physiological feature feedback, characterized in that, The method includes the following steps: The target physiological index to be adjusted is determined, and a sensory parameter space is defined. The sensory parameter space consists of the value ranges of M preset physical control parameters. The value range of each physical control parameter is mapped to a preset normalized numerical range. Define a control parameter vector, which consists of M preset physical control parameters. Each component of the control parameter vector corresponds to a physical control parameter, and the physical control parameters include multiple sensory modal control parameters. Within the sensory parameter space, N preset control parameter vectors are generated based on a preset sampling strategy. For each control parameter vector, the control parameter vector is converted into a physical control instruction through a preset hardware mapping logic and executed. After the physical control instruction corresponding to the control parameter vector is executed, the corresponding physiological signal is collected and the corresponding physiological feature vector is generated. Each component of the physiological feature vector corresponds to a target physiological indicator. Each control parameter vector and its corresponding physiological feature vector are combined into an input-output data pair, which forms a training set. A probabilistic proxy model is then constructed based on the training set. Based on the probabilistic proxy model, the training set, and the sensory parameter space, a parameter optimization iterative process is executed until the iteration stopping condition is met. When the iteration stopping condition is met, a set of non-dominated physiological feature vectors is selected from the training set. The dominance relationship is defined as follows: for any two physiological feature vectors, if every target physiological index of the first vector is not inferior to that of the second vector, and at least one index is superior to that of the second vector, then the first vector is said to dominate the second vector. The set of non-dominated physiological feature vectors in the training set refers to the set of physiological feature vectors that are not dominated by other physiological feature vectors in the training set. A physiological feature vector is selected from the set of non-dominated physiological feature vectors using a preset physiological feature vector selection strategy. The control parameter vector corresponding to the selected physiological feature vector in the training set is converted into a physical control command through a preset hardware mapping logic and executed. During execution, time windows are divided by a preset step size. The arithmetic mean of the physiological feature vectors in each sliding time window is calculated as the average physiological feature vector of the sliding time window. The current physical control command is adjusted according to the training set, the set of non-dominated physiological feature vectors, and the average physiological feature vector of the most recent sliding time window.

2. The multimodal sensory adaptive regulation method based on physiological feature feedback as described in claim 1, characterized in that, The step of converting the control parameter vector into physical control commands through preset hardware mapping logic includes: Each component of the control parameter vector is processed one by one; if the component is a visual control parameter, the component is converted into a visual control command; if the component is an auditory control parameter, the component is converted into an auditory control command; if the component is an olfactory control parameter, the component is converted into an olfactory control command.

3. The multimodal sensory adaptive regulation method based on physiological feature feedback as described in claim 1, characterized in that, After the physical control command corresponding to the control parameter vector is executed, the corresponding physiological signal is acquired and the corresponding physiological feature vector is generated, including: After the physical control command is executed, physiological signals are continuously received during the first preset adaptation buffer period, but are not included in the feature calculation. After the adaptation buffer period ends, a feature extraction sliding window of a second preset duration is opened, and the feature extraction sliding window is updated by sliding with a preset step size. The original waveform data within the feature extraction sliding window is subjected to bandpass filtering of a preset feature frequency band. The arithmetic mean of each feature data after filtering is calculated to form a physiological feature vector, where each component of the physiological feature vector is the arithmetic mean of a feature data.

4. The multimodal sensory adaptive regulation method based on physiological feature feedback as described in claim 1, characterized in that, The step of performing parameter optimization iterative process based on the probabilistic proxy model, the training set, and the sensory parameter space until the iteration stopping condition is met includes: The probabilistic proxy model is a Gaussian process regression model; Each iteration of the parameter optimization iterative process consists of four steps: Step 1: Select a set of non-dominated physiological feature vectors from the training set; Step 2: Based on the probabilistic surrogate model and the set of non-dominated physiological feature vectors, determine the control parameter vector with the largest expected hypervolume increment in the sensory parameter space. The expected hypervolume increment corresponding to the control parameter vector refers to the difference between the updated non-dominated physiological feature vector set relative to the area covered by the preset reference point and the original non-dominated physiological feature vector set relative to the area covered by the preset reference point. The updated non-dominated physiological feature vector set refers to the new non-dominated physiological feature vector set obtained by applying the probabilistic surrogate model to the control parameter vector, obtaining the predicted physiological feature vector, and adding the predicted physiological feature vector to the original non-dominated physiological feature vector set; Step 3: Convert the control parameter vector with the largest expected hypervolume increment into a physical control command through a preset hardware mapping logic and execute it to collect the corresponding physiological signal and generate the corresponding physiological feature vector; Step 4: Form an input-output data pair with the control parameter vector with the largest expected hypervolume increment and its corresponding physiological feature vector, add the input-output data pair to the training set, and reconstruct the probabilistic surrogate model based on the training set; The iteration stopping condition is when the number of iterations reaches a preset maximum value or the maximum expected supervolume increment is lower than a preset threshold.

5. The multimodal sensory adaptive regulation method based on physiological feature feedback as described in claim 1, characterized in that, The step of adjusting the current physical control command based on the training set, the non-dominated physiological feature vector set, and the average physiological feature vector of the most recent sliding time window includes: If any component of the average physiological feature vector deviates from the preset target value of the corresponding target physiological indicator by more than a preset severe deterioration threshold, then the parameter optimization iteration process is re-executed based on the probabilistic proxy model, the training set, and the sensory parameter space until the iteration stopping condition is met. When the iteration stopping condition is met, a set of non-dominated physiological feature vectors is selected from the training set, and a physiological feature vector is selected from the set of non-dominated physiological feature vectors using a preset physiological feature vector selection strategy. The control parameter vector corresponding to the selected physiological feature vector in the training set is converted into a physical control instruction through a preset hardware mapping logic, which is used as the adjusted current physical control instruction. If the average physiological feature vector is dominated by any vector in the set of non-dominated physiological feature vectors, then a physiological feature vector that dominates the average physiological feature vector is selected from the set of non-dominated physiological feature vectors using a preset physiological feature vector selection strategy, and is denoted as the target physiological feature vector. The control parameter vector corresponding to the target physiological feature vector in the training set is converted into a physical control instruction through a preset hardware mapping logic, which is then used as the adjusted current physical control instruction. If neither of the above two conditions is met, the current physical control command will remain unchanged.

6. A multimodal sensory adaptive adjustment device based on physiological feature feedback, characterized in that, The device includes: An initialization module is used to determine the target physiological index to be adjusted and to define a sensory parameter space and a control parameter vector. Within the sensory parameter space, N preset control parameter vectors are generated based on a preset sampling strategy. For each control parameter vector, the control parameter vector is converted into a physical control instruction through a preset hardware mapping logic and executed. After the physical control instruction corresponding to the control parameter vector is executed, the corresponding physiological signal is collected and the corresponding physiological feature vector is generated. Finally, each control parameter vector and its corresponding physiological feature vector are combined into an input-output data pair to form a training set. The optimization iteration module is used to construct a probabilistic proxy model based on the training set, and to execute a parameter optimization iteration process based on the probabilistic proxy model, the training set, and the sensory parameter space until the iteration stopping condition is met. The real-time monitoring and adjustment module is used to select a set of non-dominated physiological feature vectors from the training set, select a physiological feature vector from the set of non-dominated physiological feature vectors using a preset physiological feature vector selection strategy, convert the control parameter vector corresponding to the selected physiological feature vector in the training set into a physical control command through a preset hardware mapping logic and execute it, divide the time window by a preset step size during execution, calculate the arithmetic mean of the physiological feature vectors in each sliding time window as the average physiological feature vector of the sliding time window, and adjust the current physical control command according to the training set, the set of non-dominated physiological feature vectors and the average physiological feature vector of the most recent sliding time window.

7. A computer device, comprising a memory, a processor, a physiological signal acquisition module, a sensory stimulation generation module, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. The computer device as claimed in claim 7, characterized in that, The physiological signal acquisition module is used to acquire the user's raw physiological signals in real time; the physiological signal acquisition module includes at least one of an electroencephalogram (EEG) acquisition unit, an electrocardiogram (ECG) acquisition unit, a skin conductance acquisition unit, and a blood oxygen acquisition unit.

9. The computer device as claimed in claim 7, characterized in that, The sensory stimulation generation module is used to execute the physical control command to generate environmental stimulation; the sensory stimulation generation module includes at least one of a visual stimulation unit, an auditory stimulation unit, and an olfactory stimulation unit.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images