Self-adaptive personalized sleep guiding method and system and personalized sleep guiding instrument

Abstract

The invention relates to the technical field of personalized sleep guiding methods, in particular to a self-adaptive personalized sleep guiding method and system and a personalized sleep guiding instrument, and the method comprises the following steps: obtaining multi-dimensional sleep data; performing preprocessing and feature extraction on the multi-dimensional sleep data; constructing a personalized sleep model based on the preprocessing and feature extraction results; generating a sleep guide strategy according to the personalized sleep model; based on the sleep guide strategy, dynamic regulation and control of environment parameters are achieved; obtaining user feedback and updating the personalized sleep model; performing long-term effect evaluation and optimizing system parameters; by introducing a quantum heuristic multi-target award function, multiple sleep indexes can be considered in a quantum superposition state at the same time, and a test result shows that the method has remarkable advantages in multiple indexes such as sleep quality improvement (SQI), sleep time shortening rate (STDR) and deep sleep proportion increase rate (DSIR).

Description

Technical Field

[0001] The present invention relates to the technical field of personalized sleep guidance methods, and more specifically, to an adaptive personalized sleep guidance method, system and personalized sleep guidance instrument. Background Art

[0002] As people's awareness of the importance of sleep health continues to increase, personalized sleep guidance technology has developed rapidly in recent years. Existing technologies can be mainly divided into the following categories:

[0003] 1. Fixed rule-based method: This is the earliest sleep guidance method, which adjusts the sleep environment based on preset fixed rules. For example, lowering the light intensity and playing white noise at fixed times. This method is simple and direct, but lacks personalization and adaptability, and it is difficult to meet the diverse needs of different users.

[0004] 2. Single-objective optimization method: With the development of machine learning technology, some studies have begun to use single-objective reinforcement learning algorithms to optimize sleep parameters. Such methods usually select a main indicator (such as total sleep time) as the optimization target. Although they are an improvement over fixed rule methods, they often ignore the multifaceted nature of sleep quality and have difficulty optimizing multiple potentially conflicting sleep indicators at the same time.

[0005] 3. Traditional multi-objective optimization methods: In order to solve the limitations of single-objective methods, some researchers have begun to try to use multi-objective optimization algorithms. These methods can simultaneously consider multiple sleep indicators, such as sleep time, deep sleep ratio, sleep efficiency, etc. However, traditional multi-objective optimization methods often fall into local optimality when dealing with high-dimensional, nonlinear sleep problems, and it is difficult to find the global optimal solution.

[0006] Although these methods have promoted the development of sleep guidance technology to a certain extent, they still have some common technical problems:

[0007] 1. It is difficult to optimize multiple conflicting sleep indicators at the same time: Sleep is a complex physiological process involving multiple indicators that may conflict with each other. For example, increasing the proportion of deep sleep may prolong the time it takes to fall asleep. Existing methods find it difficult to find the best balance between these indicators.

[0008] 2. Insufficient modeling of environmental parameter interactions: The sleep environment involves multiple parameters such as light, sound, and temperature, which have complex interactions. Existing methods often consider these parameters independently, ignoring their mutual influence, resulting in unsatisfactory optimization results.

[0009] 3. Lack of long-term adaptability: Users’ sleep patterns and preferences may change over time, such as with seasonal changes, work pressure, etc. Existing methods are usually unable to quickly adapt to these long-term changes, resulting in a decrease in sleep effects over time.

[0010] 4. It is difficult to strike a balance between computational efficiency and optimization effect: Although some complex optimization algorithms can achieve good results, they have high computational complexity and are difficult to run in real time on resource-constrained embedded devices. Simple algorithms have high computational efficiency, but the optimization effect is often unsatisfactory.

[0011] 5. Insufficient personalization: Everyone's sleep characteristics and needs are unique, but existing methods often adopt a "one-size-fits-all" approach that makes it difficult to fully consider individual differences.

[0012] In view of the above problems, there is an urgent need for a sleep guidance method that can simultaneously handle multi-objective optimization, environmental parameter interaction, long-term adaptability, and take into account both computational efficiency and personalized needs. Summary of the invention

[0013] The purpose of the present invention is to provide an adaptive personalized sleep guidance method and control system and a personalized sleep guidance instrument to solve the problems in the prior art such as difficulty in simultaneously optimizing multiple conflicting sleep indicators, insufficient interactive modeling of environmental parameters, and lack of long-term adaptability.

[0014] The present invention provides an adaptive personalized sleep guidance method, comprising the following steps: obtaining multidimensional sleep data; preprocessing and feature extraction of the multidimensional sleep data; constructing a personalized sleep model based on the results of the preprocessing and feature extraction; generating a sleep guidance strategy based on the personalized sleep model; realizing dynamic regulation of environmental parameters based on the sleep guidance strategy; obtaining user feedback and updating the personalized sleep model;

[0015] Conduct long-term effect evaluation and optimize system parameters.

[0016] Specifically, obtaining multidimensional sleep data includes:

[0017] Acquiring physiological data, the physiological data including heart rate variability, respiratory rate, body movement and electroencephalogram data;

[0018] Acquiring environmental data, the environmental data including temperature, humidity, CO2 concentration, light intensity and color temperature, and environmental noise level;

[0019] The subjective data include a sleep quality score and a daytime mental state assessment filled in by the user.

[0020] Specifically, the preprocessing and feature extraction of multidimensional sleep data includes:

[0021] Perform wavelet transform to denoise physiological signals;

[0022] Apply a moving median filter to the environmental data to remove outliers,

[0023] Calculate time-domain indicators of heart rate variability, including SDNN, RMSSD, and pNN50;

[0024] Perform fast Fourier transform on EEG data to extract energy in the delta, theta, alpha, and beta bands;

[0025] Calculate respiratory rate variability and body movement index,

[0026] The autoencoder long short-term memory network is used to extract temporal features, where

[0027] X encoded =AE-LSTM(X raw )

[0028] Among them, X raw is the original time series data, X encoded is the encoded feature vector.

[0029] Specifically, the construction of the personalized sleep model adopts a quantum-inspired multi-objective hierarchical reinforcement learning algorithm, including:

[0030] Designing quantum-inspired multi-objective reward functions:

[0031] R=|ψ>=α|R 睡眠质量 >+β|R 效率 >+γ|R 主观感受 >

[0032] Here, |ψ> represents the quantum state of the reward, and α, β, and γ are complex amplitudes that satisfy |α| 2 +|β| 2 +|γ| 2 =1;

[0033] R 睡眠质量 =w1*f(deep sleep ratio)+w2*g(REM sleep ratio)

[0034] R 效率 =w3*h(speed of falling asleep)+w4*i(sleep duration)

[0035] R 主观感受 =w5*j(user rating)+w6*k(daytime performance)

[0036] Among them, w1 to w6 are weight coefficients, and f, g, h, i, j, and k are normalization functions;

[0037] Perform quantum-inspired weight updates:

[0038]

[0039] Among them, η is the learning rate, θ i is the quantum rotation angle, Denote the quantum state R with respect to w i The expected gradient of .

[0040] Specifically, it also includes designing a multi-layer strategy network based on fractal geometry:

[0041]

[0042]

[0043]

[0044] π I =ArgmaxQ(s,a)

[0045] Among them, π M For macro strategy, π W is a meso-strategy, π D is the micro-strategy, π I It is an instantaneous strategy;

[0047]

[0048] Among them, λ is the learning rate, w i is the weight of the lower-level strategy.

[0049] Specifically, it also includes combining quantum behavior trees and graph attention networks to model environmental parameter interactions:

[0050] H = QBAT - GAT (X, A, θ)

[0051] Among them, X is the environment parameter vector, A is the parameter relationship adjacency matrix, θ is the quantum behavior tree parameter, and H is the output environment parameter representation.

[0052] Specifically, it also includes the use of fractional Brownian motion models to capture long-term sleep pattern changes:

[0053] s t =FBM-LSTM(S t-1 , a t-1 , r t-1 , H)

[0054] Among them, FBM-LSTM is a long short-term memory network combined with fractional Brownian motion, s t is the state at time t, a t-1 is the action at the previous moment, r t-1is the reward at the previous moment, and H is the Hurst index.

[0055] Specifically, the dynamic regulation of the environmental parameters includes:

[0056] Mapping strategies to specific hardware control instructions;

[0057] Apply Bezier curve interpolation to achieve smooth parameter changes:

[0058]

[0059] Where P(t) is the parameter value at time t, P i is the control point, C(n, i) is the binomial coefficient, and the PID controller is used to adjust the parameters in real time:

[0060]

[0061] Among them, u(t) is the control signal, e(t) is the error signal, K p , K i , K d is the PID parameter.

[0062] The present invention also includes an adaptive personalized sleep guidance system, comprising:

[0063] A multi-dimensional sleep data acquisition module, used to obtain multi-dimensional sleep data;

[0064] A data preprocessing and feature extraction module, used for preprocessing and extracting features from the multi-dimensional sleep data;

[0065] A personalized sleep model building module, used to build a personalized sleep model based on the results of the preprocessing and feature extraction;

[0066] A sleep guidance strategy generating module, used for generating a sleep guidance strategy according to the personalized sleep model;

[0067] An environmental parameter dynamic control module, used to realize dynamic control of environmental parameters based on the sleep guidance strategy;

[0068] A user feedback collection and model update module, used to obtain user feedback and update the personalized sleep model;

[0069] Long-term effect evaluation and system optimization module, used to conduct long-term effect evaluation and optimize system parameters;

[0070] Among them, the personalized sleep model construction module adopts the quantum-inspired multi-objective hierarchical reinforcement learning algorithm.

[0071] The present invention also discloses a personalized sleep guidance device, comprising:

[0072] A processor, configured to execute the method;

[0073] A memory, used to store execution instructions of the method;

[0074] Communication interface, used for exchanging data with external devices;

[0075] Environmental control devices, including smart light bulbs, smart speakers and smart thermostats, for performing dynamic regulation of the environmental parameters;

[0076] The multimodal sensor includes a physiological data sensor, an environmental data sensor and a user input interface, and is used to obtain the multidimensional sleep data and user feedback.

[0077] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0078] 1. Multi-objective collaborative optimization: By introducing a quantum-inspired multi-objective reward function, the present invention can simultaneously consider multiple sleep indicators in a quantum superposition state, effectively solving the collaborative optimization problem of multiple potentially contradictory sleep indicators. Test results show that the present invention has achieved significant advantages in multiple indicators such as sleep quality improvement (SQI), shortened sleep time (STDR), and increased deep sleep ratio (DSIR).

[0079] 2. Efficient environmental parameter interaction modeling: By combining quantum behavior trees and graph attention networks, the present invention can effectively capture the complex nonlinear interactions between sleep environment parameters, improving the model's expressive power and optimization effect.

[0080] 3. Excellent long-term adaptability: By introducing fractional Brownian motion time series fusion technology, the present invention can effectively capture the long-term change trend of sleep patterns. Test results show that the present invention can adapt to changes in user sleep patterns in just 3 days, which is much faster than other methods.

[0081] 4. Good balance between computational efficiency and optimization effect: Although the present invention uses a complex algorithm, it successfully controls the computational time within an acceptable range while ensuring the optimization effect through quantum heuristic methods and hierarchical structure design. Tests show that the average computational time for each strategy update is 0.8 seconds, which is faster than ordinary multi-objective methods.

[0082] 5. Highly personalized: The present invention can fully consider the unique sleep characteristics and needs of each user through multi-dimensional sleep data collection and personalized sleep model construction. This is reflected in the high user satisfaction (8.7 / 10).

[0083] 6. Stable and reliable long-term effects: Through long-term effect evaluation and system optimization steps, the present invention can continuously optimize its performance and ensure long-term stable sleep improvement effects.

[0084] In summary, the present invention not only effectively solves multiple key problems existing in the prior art, but also shows significant advantages in terms of sleep quality improvement, user satisfaction and system adaptability. This provides a new direction for the development of personalized sleep guidance technology, and is expected to bring substantial improvement in the quality of life for users with sleep problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 It is the overall logic block diagram of the system of the present invention.

[0086] Figure 2 This is a logic block diagram of the multi-dimensional sleep data acquisition module of the present invention.

[0087] Figure 3 This is a logic block diagram of the data preprocessing and feature extraction module of the present invention.

[0088] Figure 4 The figure is a logic block diagram of the personalized sleep model building module of the present invention.

[0089] Figure 5 A logic block diagram of a sleep guidance strategy generation module of the present invention.

[0090] Figure 6 This is a logic block diagram of the environment parameter dynamic control module of the present invention.

[0091] Figure 7 This is a logic block diagram of the user feedback collection and model updating module of the present invention.

[0092] Figure 8 This is the logic block diagram of the long-term effect evaluation and system optimization module of the present invention. DETAILED DESCRIPTION

[0093] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects thereof are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0094] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0095] Please refer to Figure 1-8The present invention provides the following technical solution: an adaptive personalized sleep guidance method, comprising the following steps:

[0096] 1. Multi-dimensional sleep data collection step. In this step, the system continuously collects the user's physiological data, environmental data, and subjective feedback data through a variety of sensors. Specifically:

[0097] 1.1 Physiological data collection

[0098] The system uses the Empatica E4 wristband (sampling rate: 64Hz) to collect the user's heart rate variability (HRV) data. This high-precision HRV data can reflect the user's autonomic nervous system activity and is an important indicator for evaluating sleep quality.

[0099] At the same time, the system uses Beddit3 sleep monitor to record the user's breathing rate, body movement and sleep stage. Beddit3 is a non-contact sensor that can be placed under the mattress to accurately capture the user's breathing and body movement information through pressure changes. These data help identify the user's sleep cycle and sleep quality.

[0100] In addition, the system also collects the user's electroencephalogram (EEG) data through the MuseS headband (sampling rate: 256Hz). EEG data can provide direct information about the user's brain activity, helping to more accurately determine sleep stages and sleep quality.

[0101] 1.2 Environmental data collection, in order to fully understand the user's sleeping environment, the system uses the AwairElement air quality monitor to record environmental parameters such as temperature, humidity, and CO2 concentration every 5 minutes. These parameters have an important impact on sleep quality. For example, a higher CO2 concentration may lead to a decrease in sleep quality.

[0102] The system also uses PhilipsHue light sensors to monitor light intensity and color temperature, recording it once a minute. Light conditions have a significant impact on the body's physiological rhythm, so accurately monitoring light changes is crucial to optimizing the sleep environment.

[0103] At the same time, the system continuously records the ambient noise level through the noise sensor of the Netatmo smart indoor camera. Noise is an important factor affecting sleep quality, and continuous monitoring can help the system take noise reduction measures when necessary.

[0104] 1.3 Subjective data collection, the system has developed a customized mobile application, through which users fill in their sleep quality scores (1-10 points) and daytime mental state assessments every morning. These subjective data provide the system with user-perceived sleep quality information, which is an important supplement to objective data.

[0105] 2. Data preprocessing and feature extraction step. In this step, the system cleans and extracts features from the collected raw data to prepare for subsequent model building.

[0106] 2.1 Data cleaning, the system uses wavelet transform to remove high-frequency noise in physiological signals. The advantage of wavelet transform is that it can analyze in both time domain and frequency domain, which can effectively remove noise in the signal while retaining important details. For physiological signals such as heart rate variability and electroencephalogram, the system uses Daubechies wavelet with a decomposition level of 5 and soft threshold method for denoising.

[0107] For environmental data, the system applies a moving median filter to remove outliers. The window size is set to 5, which can effectively remove sudden outliers while retaining the overall trend of the data.

[0108] 2.2 Feature Extraction,For heart rate variability data, the system calculates the following time domain indicators:

[0109] SDNN (standard deviation of adjacent RR intervals);

[0110] RMSSD (root mean square of differences between adjacent RR intervals);

[0111] pNN50 (the proportion of adjacent RR interval differences greater than 50 ms);

[0112] These indicators reflect the balance between the sympathetic and parasympathetic nervous systems and are important parameters for assessing sleep quality.

[0113] For EEG data, the system uses fast Fourier transform (FFT) to extract the energy of the δ (0.5-4Hz), θ (4-8Hz), α (8-6)13Hz), and β (13-30Hz) bands. The energy ratio of different bands reflects the activity state of the brain and can be used to determine the stage and depth of sleep.

[0114] The system also calculates respiratory rate variability (RRV), which can reflect the regulatory ability of the autonomic nervous system and the stability of sleep. For body movement data, the system calculates the number of turns and body movement index. These indicators can reflect the restlessness of sleep. 2.3 Time series features In order to capture the time series features of the data, the system uses the autoencoder long short-term memory network (AE-LSTM) to extract time series features:

[0115] X encoded =AE-LSTM(X raw )

[0116] Among them, X raw is the original time series data, X encodedis the encoded feature vector. The structure of the AE-LSTM network is as follows:

[0117] Encoder: two layers of LSTM (128 units per layer) + one fully connected layer (64 units);

[0118] Decoder: Two layers of LSTM (128 units per layer) + one fully connected layer (same dimension as the input); this structure can effectively capture the long-term dependencies and nonlinear characteristics of the data.

[0119] 3. The personalized sleep model construction step is the core step of the present invention. The innovative quantum-inspired multi-objective hierarchical reinforcement learning algorithm (QI-MOHRL) is used to construct the personalized sleep model.

[0120] 3.1 Quantum-inspired multi-objective reward function design. The system designed a multi-objective reward function based on the concept of quantum computing:

[0121] R=|ψ>=α|R 睡眠质量 >+β|R 效率 >+γ|R 主观感受 people

[0122] Here, |ψ> represents the quantum state of the reward, and α, β, and γ are complex amplitudes that satisfy |α| 2 +|β| 2 +|γ| 2 = 1. This design allows the system to consider multiple objectives simultaneously in a quantum superposition state, overcoming the limitation of traditional methods that are difficult to balance multiple objectives.

[0123] The specific reward function is defined as follows:

[0124] R 睡眠质量 =w1·f (deep sleep ratio) + w2·g (REM sleep ratio)

[0125] R 效率 =w3·h (speed of falling asleep) + w4·i (sleep duration)

[0126] R 主观感受 =w5·j(user rating)+w6·k(daytime performance)

[0127] Among them, w1 to w6 are weight coefficients, f, g, h, i, j, k are normalization functions; k and x0 are parameters set according to experience.

[0128] The system adopts a quantum-inspired weight update mechanism:

[0129]

[0130] Where η is the learning rate (initial value is set to 0.01 and decays over time), θ i is the quantum rotation angle (initially randomly set between 0 and 2π), Denote the quantum state R with respect to w i This update mechanism introduces the concept of phase interference in quantum computing, which can explore the weight space more efficiently and avoid falling into the local optimum.

[0131] 3.2 Fractal Strategy Network,The system designs a multi-layer strategy network based on fractal geometry, including:

[0132] Macro Strategy π M : Optimize monthly sleep patterns;

[0133] Meso-strategy π W : Adjust your weekly sleep plan;

[0134] Micro Strategy π D : Optimize daily sleep parameters;

[0135] Instantaneous strategy π I : Real-time adjustment of environmental factors strategy update using fractal recursive structure:

[0136]

[0137]

[0138]

[0139] π I =ArgmaxQ(s,a)

[0140] Among them, FractalUpdate is a recursive function:

[0141]

[0142] Among them, λ is the learning rate (initial value 0.1, decaying over time), w i is the weight of the underlying strategy (calculated by the softmax function). This fractal structure allows the system to optimize the sleep strategy at different time scales and effectively handle short-term fluctuations and long-term trends.

[0143] 3.3 Quantum behavior tree environment parameter interaction modeling, the system combines quantum behavior tree and graph attention network to model environment parameter interaction:

[0144] H = QBAT - GAT (X, A, θ)

[0145] Among them, X is the environment parameter vector, A is the parameter relationship adjacency matrix, θ is the quantum behavior tree parameter, and H is the output environment parameter representation.

[0146] Forward propagation process of QBAT-GAT:

[0147] 1. Quantum behavior tree encoding: Y = QBT (X, θ)

[0148] 2. Graph attention calculation: Z = GAT(Y, A)

[0149] 3. Quantum measurement: H = Measure (Z)

[0150] QBT(X, θ) represents quantum behavior tree transformation, GAT is graph attention network, and Measure is quantum measurement operation. This method can capture complex nonlinear interactions between environmental parameters and improve the expressiveness of the model.

[0151] 3.4 Fractional Brownian motion time series fusion, the system uses the fractional Brownian motion model to capture long-term sleep pattern changes:

[0152] s t =FBM-LSTM(s t-1 , a t-1 , r t-1 , H)

[0153] Among them, FBM-LSTM is a long short-term memory network combined with fractional Brownian motion. t is the state at time t, a t-1 is the action at the previous moment, r t-1 is the reward at the previous moment, and H is the Hurst exponent (usually between 0.5 and 1, estimated based on historical data).

[0154] The update equation of FBM-LSTM is:

[0155] f t =σ(W f ·[h t-1 , x t ]+b f )

[0156] i t =σ(W i ·[h t-1 , x t ]+b i )

[0157] o t =σ(W o ·[h t-1 , x t ]+b o )

[0158]

[0159]

[0160] h t =o t ·tanh(c t )

[0161] Among them, FBM(H,t) is the fractional Brownian motion term, which is used to simulate long-term correlation. This method can effectively capture the long-term change trend of sleep patterns and improve the prediction ability of the model.

[0162] 4. Sleep guidance strategy generation step: In this step, the system generates a specific sleep guidance strategy based on the constructed personalized sleep model.

[0163] 4.1 Multi-scale strategy integration, the system integrates strategies at different time scales:

[0164] π final =ω1·π M +ω2·π W +ω3·π D +ω4·π I

[0165] Among them, ω1, ω2, ω3, and ω4 are the weights of strategies of different scales, which are dynamically adjusted according to historical effects through exponential smoothing method.

[0166] 4.2 Fuzzy logic decision system, the system uses Mamdani fuzzy inference system to convert π final Mapped to specific environment parameter adjustment instructions:

[0167] μ output =Mamdani-FIS(π final , Rule_base)

[0168] Among them, Rule_base is a predefined fuzzy rule base, which includes rules such as "IF the deep sleep ratio is low AND the sleep time is long THEN increase blue light filtering".

[0169] 4.3 Timing Constraint Optimization,The system applies Temporal Logic Programming (TLP) to ensure that the generated strategy meets the timing constraints:

[0170] Strategy final =TLP-Optimize(μ output , Temporal_constraints)

[0171] Temporal_constraints include constraints such as "light adjustment should not be performed during deep sleep", "temperature adjustment should start 30 minutes before falling asleep", etc. The TLP-Optimize function uses the Answer Set Programming technique to ensure that the generated strategy maximizes the expected reward while satisfying these temporal constraints.

[0172] 5. Dynamic control step of environmental parameters: in this step, the system controls the sleeping environment in real time according to the generated sleep guidance strategy.

[0173] 5.1 Parameter mapping, the system maps Strategy_final to specific hardware control instructions:

[0174] Philips Hue smart bulbs: adjust color temperature (2000K-6500K) and brightness (1%-100%)

[0175] Sonos One smart speaker: control volume (0-100) and audio content (such as white noise, nature sounds, etc.)

[0176] Nest Thermostat: Adjusts temperature (18°C-25°C) and humidity (40%-60%)

[0177] The mapping function uses piecewise linear interpolation to ensure that the control instructions are within the acceptable range of the hardware.

[0178] 5.2 Smooth transition algorithm, the system uses Bezier curve interpolation to achieve smooth parameter changes:

[0179]

[0180] Where P(t) is the parameter value at time t, P i is the control point, C(n, i) is the binomial coefficient. The system selects a third-order Bezier curve (n=3) that can provide sufficient smoothness while maintaining computational efficiency.

[0181] 5.3 Feedback control loop, the system uses PID (proportional-integral-differential) controller to adjust parameters in real time to ensure that the environmental state is consistent with the target:

[0182]

[0183] Among them, u(t) is the control signal, e(t) is the error signal, K p , K i , K d are PID parameters. The system uses the Ziegler-Nichols tuning method to initialize the PID parameters, and then optimizes these parameters online through the reinforcement learning method.

[0184] 6. User feedback collection and model update step: This step collects user feedback and updates the personalized sleep model accordingly.

[0185] 6.1 Multimodal Feedback Interface

[0186] Voice interaction: The system uses a BERT-based natural language processing model to analyze user voice feedback and extract emotional tendencies and keywords.

[0187] Mobile apps: Collect quantitative (e.g., 1-10 ratings) and qualitative (e.g., text descriptions) feedback through a graphical interface.

[0188] Smartwatch: records daily changes in the user's physiological indicators, such as heart rate variability, activity level, etc.

[0189] 6.2 Feedback information fusion, the system uses Dempster-Shafer evidence theory to fuse multi-source feedback:

[0190]

[0191] Among them, m(A) is the fused reliability, m i (A) is the credibility of the i-th source. The system uses the conflict factor k to adjust the credibility of different sources:

[0192]

[0193] 6.3 Online learning update, the system uses the elastic weight optimization (EWO) algorithm to update the model parameters:

[0194]

[0195] Among them, η is the learning rate (initial value 0.001, using Adam optimizer adaptive adjustment), L(θ) is the loss function (combined with mean square

[0196] Error and cross entropy), λ and p are elastic term parameters (λ = 0.01, p = 1.5). This method can quickly adapt to user feedback while maintaining model stability.

[0197] 7. Long-term effect evaluation and system optimization step: This step evaluates the long-term effect of the system and optimizes it.

[0198] 7.1 Multi-dimensional effect evaluation

[0199] Objective indicators: The system uses complex networks to analyze the topological characteristics of the sleep stage transition graph and calculates indicators such as clustering coefficient and average path length.

[0200] Subjective indicators: The system applies item response theory (IRT) to analyze user rating data and estimate potential sleep quality traits.

[0201] Physiological indicators: The system calculates the long-term trend of indicators such as HRV and EEG energy, and uses time series decomposition methods to separate trend, seasonality and random components.

[0202] 7.2 Adaptive Adjustment of System Parameters

[0203] The system uses the Lévy-based bat algorithm (Lévy-BA) to optimize system hyperparameters:

[0204]

[0205]

[0206] Among them, x i is the parameter vector, v i is the speed, f i is the pulse frequency, α is the step size control factor, A t is the average loudness, is the random step length generated by the Lévy flight. The step length of the Lévy flight follows the Lévy distribution: L(s)~s -1-β , β∈(0,2).

[0207] 7.3 Knowledge distillation and model compression, the system applies neural architecture search (NAS) and knowledge distillation technology to optimize the model structure:

[0208] L distill =λ·L CE (y,y teacher )+(1-λ)·L KL (z student , z teacher )

[0209] Among them, L CE is the cross entropy loss, L KL is the KL divergence, and λ is the balancing factor (initial value 0.5, dynamically adjusted as the training progresses). The system uses the Progressive Neural Architecture Search (PNAS) method, starting from a simple model and gradually increasing the complexity to balance model performance and computational efficiency.

[0210] Please refer to Figure 3 The present invention also discloses an adaptive personalized sleep guidance system based on the above method, and its structure includes:

[0211] The multi-dimensional sleep data collection module 1 is responsible for collecting the user's physiological data, environmental data and subjective feedback data.

[0212] The data preprocessing and feature extraction module 2 cleans and extracts features from the collected raw data.

[0213] Personalized sleep model construction module 3 uses a quantum-inspired multi-objective hierarchical reinforcement learning algorithm to build a personalized sleep model.

[0214] The sleep guidance strategy generating module 4 generates a specific sleep guidance strategy based on the personalized sleep model.

[0215] The environment parameter dynamic control module 5 controls the sleep environment parameters in real time according to the generated strategy.

[0216] The user feedback collection and model updating module 6 collects user feedback and updates the personalized sleep model.

[0217] Long-term effect evaluation and system optimization module 7 evaluates the long-term effect of the system and optimizes it.

[0218] The present invention also discloses a personalized sleep guidance device based on the system, including:

[0219] The processor executes each step of the above method.

[0220] Memory, storing execution instructions and data of the method.

[0221] Communication interface, for exchanging data with external devices.

[0222] Environmental control devices, including smart light bulbs, smart speakers, and smart thermostats, are used to perform dynamic regulation of environmental parameters.

[0223] Multimodal sensors, including physiological data sensors, environmental data sensors and user input interfaces, are used to obtain multidimensional sleep data and user feedback.

[0224] The personalized sleep guidance device of the present invention can be applied to various scenarios, such as home bedrooms, hotel rooms, long-distance flights, etc. By continuously learning and adapting to the user's sleep patterns and preferences, the system can create the best sleeping environment for each user, significantly improve sleep quality, and thus improve the user's overall health and quality of life.

[0225] Example 1 (the present invention): The adaptive personalized sleep guidance method based on quantum-inspired multi-objective hierarchical reinforcement learning described in the present invention is adopted.

[0226] Comparative Example 1: The traditional single-objective reinforcement learning method is used without considering multi-objective optimization and environmental parameter interaction.

[0227] Comparative Example 2: A multi-objective reinforcement learning method is used, but the quantum heuristic algorithm and hierarchical structure are not used.

[0228] Comparative Example 3: Using the traditional fixed-rule sleep guidance method without using machine learning technology.

[0229] In order to verify the superiority of the present invention, we conducted a series of comparative experiments. The following is a detailed comparison of Example 1 of the present invention and three comparative examples:

[0230] The test indicators and methods are as follows:

[0231] 1. Sleep quality improvement (SQI): assessed by the Pittsburgh Sleep Quality Index (PSQI) questionnaire, with a score range of 0-21, the lower the better.

[0232] 2. Sleep onset time reduction rate (STDR): measured by sleep monitoring equipment, calculated as (original sleep onset time - current sleep onset time) / original sleep onset time * 100%.

[0233] 3. Deep sleep proportion increase rate (DSIR): Through electroencephalogram (EEG) monitoring, calculate (current deep sleep proportion - original deep sleep proportion) / original deep sleep proportion * 100%.

[0234] 4. User satisfaction (US): Through questionnaire survey, the score range is 1-10, the higher the better.

[0235] 5. System Adaptability (SA): Measures the time it takes the system to adapt to changes in the user's sleep patterns (in days). The less the better.

[0236] 6. Computational efficiency (CE): The average computation time for each policy update, in seconds. The less the better.

[0237] The test results are shown in the following table:

[0238] index Example 1 Comparative Example 1 Comparative Example 2 Comparative Example 3 SQI -5.2 -2.8 -3.5 -1.9 STDR 35 18 25 12 DSIR 28 15 20 10 US 8.7 6.5 7.2 5.8 SA 375 14 1 0 CE 0.8 0.5 1.2 0.2

[0239] The analysis and explanation of the above test results are as follows:

[0240] 1. Improvement in sleep quality (SQI): The PSQI of Example 1 was reduced by 5.2 points, which was much better than other methods. This shows that the method of the present invention can improve sleep quality more comprehensively, probably because it takes into account the coordinated optimization of multiple sleep indicators.

[0241] 2. Sleeping time reduction rate (STDR): Example 1 shortened the sleeping time by 35%, which was significantly better than other methods. This may be due to the hierarchical strategy structure of the present invention, which can more accurately regulate the environmental parameters before falling asleep.

[0242] 3. Deep sleep ratio increase rate (DSIR): Example 1 increased the deep sleep ratio by 28%, which is a significant advantage. This may be because the quantum heuristic algorithm can better balance the needs of different sleep stages.

[0243] 4. User satisfaction (US): Example 1 received a high score of 8.7, reflecting the high recognition of the system by users, which may be due to the high personalization and adaptability of the system.

[0244] 5. System adaptability (SA): Example 1 can adapt to the changes in the user's sleep pattern in just 3 days, which is much faster than other methods. This highlights the advantage of the present invention in dealing with long-term temporal dependencies, which may be due to the fractional Brownian motion time series fusion technology.

[0245] 6. Computational efficiency (CE): The computation time of Example 1 is 0.8 seconds, which is slower than the simple method (Comparative Examples 1 and 3), but it is acceptable considering its complexity and effect. Moreover, it is faster than the ordinary multi-objective method (Comparative Example 2), probably because the quantum heuristic algorithm improves the optimization efficiency.

[0246] The method of the present invention shows obvious advantages in all key indicators. It can not only significantly improve sleep quality, shorten the time to fall asleep, and increase the proportion of deep sleep, but also achieve high user satisfaction. It is particularly noteworthy that this method shows excellent system adaptability and can quickly adapt to changes in user sleep patterns. Although there is a slight sacrifice in computational efficiency, this trade-off is worth it considering the significant improvement in effect it brings.

[0247] These results fully demonstrate the innovation of the present invention, namely the superiority of the quantum-inspired multi-objective hierarchical reinforcement learning algorithm. It can effectively handle the multi-objective optimization problem in sleep guidance, while improving the adaptability and optimization efficiency of the system through the hierarchical structure and quantum-inspired algorithm. In addition, the results also reflect the advantages of the present invention in dealing with long-term temporal dependencies and environmental parameter interactions, which may be due to innovative technologies such as fractional Brownian motion time series fusion and quantum behavior tree environmental parameter interaction modeling.

[0248] Based on the above analysis, Example 1 can be considered as the best example, which is significantly better than other methods in most key indicators, especially in terms of sleep quality improvement and system adaptability.

[0249] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An adaptive personalized sleep guidance method, characterized in that: The following steps are involved: Obtain multi-dimensional sleep data; Preprocessing and feature extraction are performed on the multidimensional sleep data; and based on the results of the preprocessing and feature extraction, a personalized sleep model is constructed; generating a sleep guidance strategy according to the personalized sleep model; Based on the sleep guidance strategy, dynamic regulation of environmental parameters is achieved; user feedback is obtained and the personalized sleep model is updated; long-term effect evaluation is performed and system parameters are optimized.

2. The method according to claim 1, characterized in that: The obtaining of multi-dimensional sleep data comprises: Acquiring physiological data, the physiological data including heart rate variability, respiratory rate, body movement and electroencephalogram data; Acquiring environmental data, the environmental data including temperature, humidity, CO2 concentration, light intensity and color temperature, and environmental noise level; The subjective data include a sleep quality score and a daytime mental state assessment filled in by the user.

3. The method according to claim 1, characterized in that: The preprocessing and feature extraction of multi-dimensional sleep data includes: Perform wavelet transform to denoise physiological signals; Apply a moving median filter to the environmental data to remove outliers, Calculate time-domain indicators of heart rate variability, including SDNN, RMSSD, and pNN50; Perform fast Fourier transform on EEG data to extract energy in the delta, theta, alpha, and beta bands; Calculate respiratory rate variability and body movement index, The autoencoder long short-term memory network is used to extract temporal features, where X encoded =AE-LSTM(X raw ) Among them, X raw is the original time series data, X encoded is the encoded feature vector.

4. The method according to claim 1, characterized in that: The personalized sleep model is constructed by using a quantum-inspired multi-objective hierarchical reinforcement learning algorithm, including: Designing quantum-inspired multi-objective reward functions: R=|ψ>=α|R 睡眠质量 >+β|R 效率 >+γ|R 主观感受 > Here, |ψ> represents the quantum state of the reward, and α, β, and γ are complex amplitudes that satisfy |α| 2 +|β| 2 +|γ| 2 =1; R 睡眠质量 =w1*f(deep sleep ratio)+w2*g(REM sleep ratio) R 效率 =w3*h(speed of falling asleep)+w4*i(sleep duration) R 主观感受 =w5*j(user rating)+w6*k(daytime performance) Among them, w1 to w6 are weight coefficients, and f, g, h, i, j, and k are normalization functions; Perform quantum-inspired weight updates: Among them, η is the learning rate, θ i is the quantum rotation angle, Denotes the quantum state R with respect to w i The expected gradient of .

5. The method according to claim 4, characterized in that It also includes designing multi-layer policy networks based on fractal geometry: p I =ArgmaxQ(s,a) Among them, π M For macro strategy, π W is a meso-strategy, π D is the micro-strategy, π I It is an instantaneous strategy; Among them, λ is the learning rate, w i is the weight of the lower-level strategy.

6. The method according to claim 4, characterized in that It also includes combining quantum behavior trees and graph attention networks to model environmental parameter interactions: H=QBAT-GAT(X,A,θ) Among them, X is the environment parameter vector, A is the parameter relationship adjacency matrix, θ is the quantum behavior tree parameter, and H is the output environment parameter representation.

7. The method according to claim 4, characterized in that It also includes the use of fractional Brownian motion models to capture long-term sleep pattern changes: s t =FBM-LSTM(s t-1 ,a t-1 ,r t-1 ,H) Among them, FBM-LSTM is a long short-term memory network combined with fractional Brownian motion, s t is the state at time t, a t-1 is the action at the previous moment, r t-1 is the reward at the previous moment, and H is the Hurst index.

8. The method according to claim 1, characterized in that The dynamic regulation of the environmental parameters includes: Mapping strategies to specific hardware control instructions; Apply Bezier curve interpolation to achieve smooth parameter changes: Where P(t) is the parameter value at time t, P i is the control point, C(n,i) is the binomial coefficient, and the PID controller is used to adjust the parameters in real time: Among them, u(t) is the control signal, e(t) is the error signal, K p ,K i , K d is the PID parameter.

9. Adaptive personalized sleep guidance system, characterized in that: include: A multi-dimensional sleep data acquisition module, used to obtain multi-dimensional sleep data; A data preprocessing and feature extraction module, used for preprocessing and extracting features from the multi-dimensional sleep data; A personalized sleep model building module, used to build a personalized sleep model based on the results of the preprocessing and feature extraction; A sleep guidance strategy generating module, used for generating a sleep guidance strategy according to the personalized sleep model; An environmental parameter dynamic control module, used to realize dynamic control of environmental parameters based on the sleep guidance strategy; A user feedback collection and model update module, used to obtain user feedback and update the personalized sleep model; Long-term effect evaluation and system optimization module, used to conduct long-term effect evaluation and optimize system parameters; Wherein, the personalized sleep model construction module adopts the quantum-inspired multi-objective hierarchical reinforcement learning algorithm as described in any one of claims 4 to 7.

10. Personalized sleep guidance device, characterized in that: include: A processor, configured to execute the method according to any one of claims 1 to 8; A memory, used to store execution instructions of the method; Communication interface, used for exchanging data with external devices; Environmental control devices, including smart light bulbs, smart speakers and smart thermostats, for performing dynamic regulation of the environmental parameters; The multimodal sensor includes a physiological data sensor, an environmental data sensor and a user input interface, and is used to obtain the multidimensional sleep data and user feedback.

Images