Smart home environment data fusion perception and optimization method

Abstract

The application discloses a kind of smart home environment data fusion perception and optimization method, it is related to smart home environment control technical field, including the following steps: by the distributed sensing array acquisition multi-generation physiological parameter data, based on the multi-generation physiological parameter data statistics in preset historical period establishes generation physiological characteristic benchmark atlas, based on each generation environmental demand parameter vector calculates generation environmental demand conflict measure, according to conflict measure distribution fusion weight, according to environmental physical parameter variation trend and heat mass transfer correlation, combine generation sensitivity matrix and construct multi-objective optimization function;By calculating the included angle cosine distance between each generation environmental demand parameter vector as conflict measure, according to conflict measure distribution fusion weight, carry out multi-level fusion in data layer, characteristic layer, decision layer to feature vector sequence, generate comprehensive environment state evaluation result.

Description

Technical Field

[0001] This invention belongs to the field of smart home environment control technology, specifically a smart home environment data fusion perception and optimization method. Background Technology

[0002] With the rapid development of smart home technology, home environment control systems have gradually evolved from simple temperature regulation to intelligent systems with multi-parameter coordinated control.

[0003] Despite the progress made by existing technologies in their respective fields, the following technical shortcomings still exist in the scenario of multi-generational coexistence home environment control:

[0004] Lacking generation-specific perception capabilities, existing systems are not specifically designed to address the physiological differences among different generations such as the elderly, adults, and children, making it impossible to identify the identities of different generations and meet the differentiated needs of each generation.

[0005] The data fusion is too simple. Existing systems mostly use a single-level data fusion method, which cannot simultaneously eliminate differences in units, improve inter-subjective correlation features, and resolve decision conflicts, resulting in limited fusion effects and low accuracy in environmental pattern recognition. Summary of the Invention

[0006] The technical objective of this invention is to provide a multi-generational coexistence home environment control system and control method, which can simultaneously sense the physiological parameters, environmental parameters, and intergenerational physiological coupling effects of multiple generations of members. Through multi-level progressive data fusion and multi-objective optimization based on the intergenerational sensitivity matrix, it achieves a differentiated balance of the comfort needs of different generations of members, and ensures the secure transmission of physiological parameter data through a differential privacy mechanism.

[0007] This paper provides a method for data fusion, perception, and optimization in smart home environments to address the problems mentioned in the background section.

[0008] A method for data fusion, perception, and optimization of smart home environments includes the following steps:

[0009] S1. Collect multi-generational physiological parameter data, environmental physical parameter data, intergenerational activity area data, and intergenerational physiological coupling effect data through the distributed sensor array;

[0010] The distributed sensing array includes a millimeter-wave radar array and a depth vision sensor array. The millimeter-wave radar array is a distributed sensing array composed of a set number of 60GHz frequency-modulated continuous wave radar nodes. The depth vision sensor array is a distributed vision array composed of a set number of time-of-flight depth cameras.

[0011] The multigenerational physiological parameter data includes non-contact heart rate, respiratory rate, and body surface temperature. The non-contact heart rate is the heart rate value obtained by converting the radar echo phase change frequency caused by micro-movement of the chest cavity by millimeter-wave radar. The respiratory rate is the respiratory rate value obtained by converting the radar echo amplitude modulation period caused by abdominal undulation by millimeter-wave radar. The body surface temperature is the temperature value obtained by collecting the temperature distribution of the facial area by an infrared thermal imaging sensor and calculating the regional weighted average.

[0012] The environmental physical parameter data include indoor temperature, indoor humidity, light intensity, air quality index, and environmental noise level;

[0013] The intergenerational activity area data consists of the location coordinates and dwell time of each generation member in the indoor space. The location coordinates are obtained by mapping the center point of the human body contour collected by the depth vision sensor to the indoor plane coordinate system.

[0014] The intergenerational physiological coupling effect data are the correlation coefficient of heart rate variability and respiratory synchronization rate among members of different generations in the same space.

[0015] S2. Based on the statistical analysis of multi-generational physiological parameter data within a preset historical period, establish a benchmark map of intergenerational physiological characteristics. Normalize the current multi-generational physiological parameter data relative to the benchmark map of intergenerational physiological characteristics by offset. Asynchronously align the multi-source data streams with the intergenerational activity rhythm period as the window length, and extract time-frequency features to construct a feature vector sequence.

[0016] The intergenerational physiological characteristic baseline map is a statistical baseline interval characterizing the normal fluctuation range of various physiological parameters of each generation member, and is used as the physiological basis for judging environmental regulation needs.

[0017] The intergenerational activity rhythm cycle is the arithmetic mean of the start and end time intervals of activities of each generation member within a preset historical cycle.

[0018] S3. Calculate the intergenerational environmental demand conflict measure based on the intergenerational environmental demand parameter vectors, allocate fusion weights according to the conflict measure, and perform multi-level fusion of feature vector sequences in the data layer, feature layer and decision layer to generate comprehensive environmental status assessment results. Use the comprehensive environmental status assessment results to predict the changing trend of environmental physical parameters through a time series prediction network, and model the heat and mass transfer correlation between different intergenerational activity areas through a graph structure neural network.

[0019] The conflict metric is the cosine distance between the environmental demand parameter vectors of different generations, and the environmental demand parameter vector is a multidimensional vector composed of the optimal demand values ​​of each generation for multiple environmental physical parameters.

[0020] The heat and mass transfer correlation refers to the intensity of physical transfer of heat, humidity and airflow between different generational activity areas through the building ventilation structure.

[0021] S4. Based on the changing trends of environmental physical parameters and the correlation of heat and mass transfer, a multi-objective optimization function is constructed by combining the intergenerational sensitivity matrix. An environmental regulation control command is generated by solving the strategy optimization network and sent to the environmental regulation actuator. The environmental regulation actuator changes the indoor physical environmental parameters according to the environmental regulation control command.

[0022] The environmental control actuators include air conditioning units, fresh air units, lighting actuators, and curtain actuators;

[0023] The environmental regulation control commands include temperature setpoint, fan speed setting, light intensity value, and curtain opening / closing degree;

[0024] The multi-objective optimization function includes an intergenerational comfort deviation term, a system energy consumption term, and an equipment switching frequency term. The intergenerational comfort deviation term is the weighted sum of squares of the differences between the actual environmental physical parameters of each generation and the corresponding historical preferred environmental physical parameters of the generation. The system energy consumption term is the total power consumption of the current environmental regulation actuators. The equipment switching frequency term is the number of times the environmental regulation actuators start and stop per unit time.

[0025] The generational sensitivity matrix is ​​obtained by statistically analyzing the response slope of historical physiological parameter data of each generation to changes in environmental physical parameters. The response slope is the linear fitting coefficient between the changes in physiological parameters of each generation and the changes in environmental physical parameters.

[0026] S5. Collect feedback data of indoor physical environment parameters and multi-generational physiological parameters after execution, calculate the deviation between the actual value of physical environment parameters and the target set value. When the deviation of any generation exceeds the preset multiple of the statistical standard deviation of the historical deviation of that generation, or when the weighted sum of the deviations of a set number of generations exceeds the preset multiple of the statistical standard deviation of the global historical weighted deviation, return to step S1.

[0027] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects:

[0028] By calculating the cosine distance between the environmental demand parameter vectors of different generations as a conflict metric, and assigning fusion weights based on the conflict metric, the feature vector sequences are fused at multiple levels—data layer, feature layer, and decision layer—to generate a comprehensive environmental state assessment result. Simultaneously, a temporal prediction network is used to predict the changing trends of environmental physical parameters, and a graph structure neural network is used to model the heat and mass transfer relationships between different generational activity areas.

[0029] Based on statistical analysis of multi-generational physiological parameter data within a preset historical period, a benchmark map of intergenerational physiological characteristics is established. The current physiological parameter data is normalized relative to the benchmark map by offset. Asynchronous window alignment of the multi-source data streams is performed using the intergenerational activity rhythm cycle as the window length. Time-frequency features are extracted and a feature vector sequence is constructed through short-time Fourier transform, multi-scale wavelet decomposition, and statistical moments of various orders. This scheme effectively eliminates individual differences and achieves accurate identification of environmental disturbance characteristics corresponding to different intergenerational activity patterns. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the system framework of the present invention. Detailed Implementation

[0031] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0032] Please see Figure 1 This application provides a method for data fusion sensing and optimization of smart home environment, including the following steps:

[0033] S1. Collect multi-generational physiological parameter data, environmental physical parameter data, intergenerational activity area data, and intergenerational physiological coupling effect data through the distributed sensor array;

[0034] The distributed sensing array includes a millimeter-wave radar array and a depth vision sensor array. The millimeter-wave radar array is a distributed sensing array composed of a set number of 60GHz frequency-modulated continuous wave radar nodes. The depth vision sensor array is a distributed vision array composed of a set number of time-of-flight depth cameras.

[0035] The multigenerational physiological parameter data includes non-contact heart rate, respiratory rate, and body surface temperature. The non-contact heart rate is the heart rate value obtained by converting the radar echo phase change frequency caused by micro-movement of the chest cavity by millimeter-wave radar. The respiratory rate is the respiratory rate value obtained by converting the radar echo amplitude modulation period caused by abdominal undulation by millimeter-wave radar. The body surface temperature is the temperature value obtained by collecting the temperature distribution of the facial area by an infrared thermal imaging sensor and calculating the regional weighted average.

[0036] The environmental physical parameter data include indoor temperature, indoor humidity, light intensity, air quality index, and environmental noise level;

[0037] The intergenerational activity area data consists of the location coordinates and dwell time of each generation member in the indoor space. The location coordinates are obtained by mapping the center point of the human body contour collected by the depth vision sensor to the indoor plane coordinate system.

[0038] The intergenerational physiological coupling effect data are the correlation coefficient of heart rate variability and respiratory synchronization rate among members of different generations in the same space.

[0039] S2. Based on the statistical analysis of multi-generational physiological parameter data within a preset historical period, establish a benchmark map of intergenerational physiological characteristics. Normalize the current multi-generational physiological parameter data relative to the benchmark map of intergenerational physiological characteristics by offset. Asynchronously align the multi-source data streams with the intergenerational activity rhythm period as the window length, and extract time-frequency features to construct a feature vector sequence.

[0040] The intergenerational physiological characteristic baseline map is a statistical baseline interval characterizing the normal fluctuation range of various physiological parameters of each generation member, and is used as the physiological basis for judging environmental regulation needs.

[0041] The intergenerational activity rhythm cycle is the arithmetic mean of the start and end time intervals of activities of each generation member within a preset historical cycle.

[0042] S3. Calculate the intergenerational environmental demand conflict measure based on the intergenerational environmental demand parameter vectors, allocate fusion weights according to the conflict measure, and perform multi-level fusion of feature vector sequences in the data layer, feature layer and decision layer to generate comprehensive environmental status assessment results. Use the comprehensive environmental status assessment results to predict the changing trend of environmental physical parameters through a time series prediction network, and model the heat and mass transfer correlation between different intergenerational activity areas through a graph structure neural network.

[0043] The conflict metric is the cosine distance between the environmental demand parameter vectors of different generations, and the environmental demand parameter vector is a multidimensional vector composed of the optimal demand values ​​of each generation for multiple environmental physical parameters.

[0044] The heat and mass transfer correlation refers to the intensity of physical transfer of heat, humidity and airflow between different generational activity areas through the building ventilation structure.

[0045] S4. Based on the changing trends of environmental physical parameters and the correlation of heat and mass transfer, a multi-objective optimization function is constructed by combining the intergenerational sensitivity matrix. An environmental regulation control command is generated by solving the strategy optimization network and sent to the environmental regulation actuator. The environmental regulation actuator changes the indoor physical environmental parameters according to the environmental regulation control command.

[0046] The environmental control actuators include air conditioning units, fresh air units, lighting actuators, and curtain actuators;

[0047] The environmental regulation control commands include temperature setpoint, fan speed setting, light intensity value, and curtain opening / closing degree;

[0048] The multi-objective optimization function includes an intergenerational comfort deviation term, a system energy consumption term, and an equipment switching frequency term. The intergenerational comfort deviation term is the weighted sum of squares of the differences between the actual environmental physical parameters of each generation and the corresponding historical preferred environmental physical parameters of the generation. The system energy consumption term is the total power consumption of the current environmental regulation actuators. The equipment switching frequency term is the number of times the environmental regulation actuators start and stop per unit time.

[0049] The generational sensitivity matrix is ​​obtained by statistically analyzing the response slope of historical physiological parameter data of each generation to changes in environmental physical parameters. The response slope is the linear fitting coefficient between the changes in physiological parameters of each generation and the changes in environmental physical parameters.

[0050] S5. Collect feedback data of indoor physical environment parameters and multi-generational physiological parameters after execution, calculate the deviation between the actual value of physical environment parameters and the target set value. When the deviation of any generation exceeds the preset multiple of the statistical standard deviation of the historical deviation of that generation, or when the weighted sum of the deviations of a set number of generations exceeds the preset multiple of the statistical standard deviation of the global historical weighted deviation, return to step S1.

[0051] As an optional embodiment, the collection and use of intergenerational physiological coupling effect data in step S1 includes:

[0052] Simultaneously collect heart rate variability and respiratory rate sequences of different generations of members in the same enclosed indoor space using a millimeter-wave radar array.

[0053] The heart rate variability sequence is a time series data composed of the interval differences between consecutive heartbeats. The heart rate is obtained by converting the frequency of radar echo phase changes caused by micro-movements in the chest cavity by millimeter-wave radar. The interval differences between consecutive heartbeats are then calculated.

[0054] The respiratory rate sequence is a time series data composed of respiratory cycles, which is obtained by converting the radar echo amplitude modulation period caused by abdominal undulation detected by millimeter-wave radar.

[0055] Calculate the correlation coefficient of heart rate variability: Calculate the ratio of the product of the covariance and the standard deviation of the heart rate variability sequences of the i-th generation member and the j-th generation member to obtain the correlation coefficient of heart rate variability, which characterizes the degree of synchronization of heart rate fluctuations between the two generations.

[0056] Calculate the respiratory synchronization rate: Perform phase synchronization analysis on the respiratory rate sequences of the i-th generation member and the j-th generation member to obtain the respiratory synchronization rate, which characterizes the degree of synchronization of the respiratory rhythms of the two generations. The phase synchronization analysis is a signal processing operation that calculates the complex exponential average of the phase difference after obtaining the instantaneous phase by performing Hilbert transform on the respiratory rate sequence.

[0057] In step S3, the weight allocation of each generation region in the environmental status assessment is adjusted according to the intergenerational physiological coupling effect data. The adjustment is as follows: when the heart rate variability correlation coefficient is greater than the first preset threshold and the respiratory synchronization rate is greater than the second preset threshold, the two generation members are determined to be in a strong coupling state, and the weight allocation ratio of the strong coupling state generation region in the environmental status assessment is increased to a preset multiple of the original weight allocation ratio.

[0058] The original weight allocation ratio is the weight allocation ratio value allocated in step S3 through conflict metric sum exponential calculation and normalization.

[0059] The first preset threshold and the second preset threshold are determined through experimental verification. The experimental verification involves collecting heart rate variability correlation coefficient and respiratory synchronization rate data of multiple intergenerational combinations in a controlled environment, and selecting the threshold parameters after statistically analyzing the boundary values ​​between strong coupling and weak coupling states.

[0060] As an optional embodiment, the determination of intergenerational membership and adaptive sampling in step S1 includes:

[0061] The millimeter-wave radar array transmits a frequency-modulated continuous wave signal in the 60GHz band, and after receiving the echo signal, it performs frequency mixing to obtain an intermediate frequency signal. The intermediate frequency signal is a low-frequency signal composed of the frequency difference between the transmitted signal and the echo signal.

[0062] The intermediate frequency signal is subjected to distance-dimensional fast Fourier transform and Doppler-dimensional fast Fourier transform to extract the Doppler frequency shift information of the human target as gait periodic features. The gait periodic features are the reciprocals of the dominant frequency components of the power spectrum of the velocity signal during human walking.

[0063] Indoor depth image sequences are acquired at a rate of 30 frames per second using a depth vision sensor array. The height ratio, width-to-height ratio, and shoulder-to-hip ratio of the human body contour are extracted as body contour proportion features. The height ratio is the ratio of the height of the human body contour to the preset average human body height. The width-to-height ratio is the ratio of the width of the human body contour to the height. The shoulder-to-hip ratio is the ratio of the width of the shoulder to the width of the hip of the human body contour.

[0064] Gait cycle features and body shape proportion features are concatenated to form an intergenerational biological feature vector, which is input into a classification model based on radial basis function kernel function, and outputs intergenerational identity as elderly member, adult member or minor member. The radial basis function kernel function is a kernel function that takes the Euclidean distance between feature vectors as input and outputs a similarity value through exponential decay operation.

[0065] The classification model determines the regularization parameters and kernel function parameters through parameter optimization experiments. The parameter optimization experiments are an experimental process of evaluating the model accuracy and selecting the optimal parameter combination within a preset parameter range through cross-validation.

[0066] The sampling frequency of the sensor nodes corresponding to each generation's activity area in the distributed sensing array is adaptively adjusted according to the rate of change of activity intensity of each generation member. The rate of change of activity intensity is determined by the variance of the displacement of position coordinates per unit time, and the displacement of position coordinates is the Euclidean distance between position coordinates at adjacent sampling times.

[0067] When the rate of change in activity intensity is greater than the sum of the mean and standard deviation of the historical rate of change in activity intensity for that generation, the acquisition frequency of the corresponding sensor nodes is increased to twice the base frequency.

[0068] When the rate of change of activity intensity is less than or equal to the sum of the mean and standard deviation of the historical rate of change of activity intensity in that generation, the acquisition frequency of the corresponding sensor nodes in the region is reduced to 0.5 times the base frequency.

[0069] The fundamental frequency is the lowest effective frequency required for sampling physiological parameter data across generations, determined by the Nyquist sampling theorem based on the highest frequency component of the physiological signal.

[0070] As an optional embodiment, the specific implementation of establishing the intergenerational physiological characteristic benchmark map and extracting time-frequency features in step S2 includes:

[0071] The physiological parameter data of each generation member are statistically analyzed within a continuous preset historical period. The arithmetic mean of each physiological parameter data of each generation is calculated as the distribution mean. The square root of the arithmetic mean of the squares of the deviations of each physiological parameter data of each generation from the distribution mean is calculated as the distribution variance. The distribution mean is used as the baseline center and the distribution variance of the preset multiple is used as the baseline fluctuation range to establish a benchmark map of intergenerational physiological characteristics.

[0072] The intergenerational physiological characteristic baseline map is a statistical baseline interval characterizing the normal fluctuation range of various physiological parameters of each generation member, and is used as the physiological basis for judging environmental regulation needs.

[0073] The current multigenerational physiological parameter data is normalized relative to the corresponding generational physiological characteristic baseline map by offset. The offset normalization is a data transformation process that subtracts the center value of the corresponding generational physiological characteristic baseline map from the current physiological parameter data and then divides it by half the width of the fluctuation range of the corresponding generational physiological characteristic baseline map.

[0074] The intergenerational activity rhythm cycle is obtained by statistically analyzing the historical activity records of each generation member. The intergenerational activity rhythm cycle is the arithmetic mean of the start and end time intervals of the activities of each generation member within a continuous preset historical cycle. The start and end times of the activities are the start and end times of the continuous period in which the change in the position coordinates of indoor personnel exceeds a preset displacement threshold.

[0075] Asynchronous window alignment is performed on multi-source data streams using the intergenerational activity rhythm cycle as the window length, and time-frequency features are extracted to construct a feature vector sequence. The time-frequency features are used to identify environmental disturbance features corresponding to different intergenerational activity patterns.

[0076] The time-frequency features include:

[0077] The frequency domain features are extracted by short-time Fourier transform, which uses a Hanning window with a window length of one-quarter of the intergenerational activity rhythm period and a frame shift of one-half of the window length.

[0078] The time-domain local features are extracted through multi-scale wavelet decomposition, which uses compactly supported orthogonal wavelet basis functions with fourth-order vanishing moments and has a decomposition level of 5.

[0079] The data distribution characteristics extracted by statistical moments of various orders include the first-order raw moment, the second-order central moment, the third-order standardized moment, and the fourth-order standardized moment. The first-order raw moment is the arithmetic mean of the data, the second-order central moment is the arithmetic mean of the squared deviations of the data from the mean, the third-order standardized moment is the third moment after data standardization, and the fourth-order standardized moment is the fourth moment after data standardization minus 3.

[0080] As an optional embodiment, the specific implementation of the multi-level fusion in step S3 includes:

[0081] The data layer adopts a weighted average fusion based on fusion weights. The weighted average fusion is the sum of the products of each generation's original data stream and the corresponding fusion weight, which is used to eliminate the dimensional differences between different generations of data streams.

[0082] The feature layer employs a fusion based on an intergenerational attention mechanism, which calculates intergenerational attention scores and performs normalization to obtain attention-weighted features, which are used to extract correlation features between different intergenerational data streams.

[0083] The decision-making level employs evidence fusion based on conflict metric correction, which includes adjusting the basic probability allocation values ​​of each generation's evidence according to the conflict metric, in order to resolve conflicts between decision outcomes of different generations.

[0084] The progressive logic of the multi-level fusion is as follows: after the data layer fusion eliminates the difference in dimensions, it is input to the feature layer fusion to extract related features; after the feature layer fusion outputs the related features, it is input to the decision layer fusion to resolve decision conflicts; and the three-level fusion outputs the comprehensive environmental state assessment result, which is used to generate subsequent environmental regulation and control instructions.

[0085] As an optional embodiment, the edge weights of the graph structure neural network in step S3 are determined by the ratio of the overlapping area of ​​the intergenerational activity regions to the cross-sectional area of ​​the airflow channels between the connected regions.

[0086] The overlapping area is the planar area of ​​the part where the two intergenerational activity areas intersect on the indoor space plan.

[0087] The cross-sectional area of ​​the airflow channel is obtained by measuring the following steps:

[0088] For ventilation ducts, measure the inner diameter of the duct and calculate the cross-sectional area, which is the product of pi and the square of half the inner diameter of the duct.

[0089] For door and window openings, measure the width and height of the opening and calculate the cross-sectional area, which is the product of the opening width and the opening height.

[0090] For regions without physical connections, the edge weights are set to a minimum positive number, which is the lower limit of the machine's floating-point precision.

[0091] The edge weights are used to model the intensity of physical transfer of heat, humidity, and airflow through the building's ventilation structure between different generational activity areas.

[0092] As an optional embodiment, the construction of the intergenerational sensitivity matrix in step S4 includes:

[0093] The other four environmental physical parameters are fixed at the median value. Only one environmental physical parameter is changed, and the corresponding physiological parameter change is recorded. The median value is the arithmetic mean of the preset effective range of the environmental physical parameters.

[0094] The response slope of each generation of physiological parameter data to changes in environmental physical parameters is calculated. The response slope is the linear fitting coefficient of the changes in physiological parameters and environmental physical parameters in each generation. The linear fitting coefficient is the regression coefficient obtained by fitting the linear relationship between the changes in physiological parameters and environmental physical parameters.

[0095] After normalizing the response slopes, an intergenerational sensitivity matrix is ​​constructed. The normalization is an operation in which the absolute value of each response slope is divided by the sum of the absolute values ​​of all response slopes.

[0096] The intergenerational sensitivity matrix is ​​used to quantify the physiological sensitivity of different generations to temperature, humidity, light and noise, and serves as the basis for the weighting of the intergenerational comfort deviation term in the multi-objective optimization function.

[0097] As an optional embodiment, the solution of the multi-objective optimization function in step S5 adopts a strategy optimization network based on dynamic adjustment of generational priorities.

[0098] The policy optimization network is a deep neural network structure that includes a policy generation network and a value evaluation network. The policy generation network is a deep neural network that includes an input layer, a hidden layer, and an output layer. The hidden layer uses a non-linear activation function, and the output layer uses a bounded output activation function to generate control commands for air conditioning units, fresh air units, lighting drivers, and curtain drivers.

[0099] The value evaluation network is a deep neural network containing an input layer, a hidden layer, and an output layer, used to evaluate the expected returns of the adjustment strategy.

[0100] The intergenerational priority is dynamically updated based on the satisfaction statistics of historical adjustment feedback of each generation. The satisfaction is the ratio of the absolute value of the Z-score offset of the physiological index after adjustment to the absolute value of the Z-score offset of the physiological index before adjustment. When the absolute value of the Z-score offset before adjustment is zero, the satisfaction is the absolute value of the Z-score offset after adjustment.

[0101] As an optional implementation, this includes: establishing an intergenerational environmental preference model.

[0102] By accumulating data over a continuously preset period, the distribution of environmental physical parameter preferences of different generations in different activity scenarios is statistically analyzed. The activity scenarios include sleep scenarios, dining scenarios, learning scenarios, leisure scenarios, and sports scenarios. The sleep scenario is the time spent resting in bed at night. The dining scenario is the time spent eating in a restaurant or kitchen. The learning scenario is the time spent reading and writing in a study or at a desk. The leisure scenario is the time spent watching and listening to entertainment in the living room. The sports scenario is the time spent engaging in physical activities indoors.

[0103] Calculate the expected value and variance of the environmental physical parameter preference distribution, where the expected value is the arithmetic mean of historical environmental physical parameter setting data, and the variance is the arithmetic mean of the squared deviations between the historical environmental physical parameter setting data and the expected value.

[0104] In step S4, a robust environmental setting value is generated by combining the expected value and variance of the environmental physical parameter preference distribution. The robust environmental setting value is the target setting value of the environmental physical parameters that minimizes the intergenerational comfort deviation term within a preset confidence interval.

[0105] The lower limit of the preset confidence interval is the expected value minus the product of the statistical critical value and the standard error. The upper limit of the preset confidence interval is the expected value plus the product of the statistical critical value and the standard error. The statistical critical value is the t-distribution critical value obtained by looking up the table based on the sample degrees of freedom. The standard error is the square root of the quotient of the variance and the sample size.

[0106] As an optional embodiment, it also includes:

[0107] The real-time processing of physiological parameter data involving multiple generations is deployed on a local edge computing node. The local edge computing node is an embedded computing platform that includes an ARM processor, a graphics processing acceleration unit, and a memory storage unit of no less than 32GB. The local edge computing node is directly connected to the distributed sensor array and the environmental regulation actuator through a wired or wireless communication interface.

[0108] The environment model reasoning and global optimization solution are deployed on a remote cloud server.

[0109] Before transmitting data between local edge computing nodes and remote cloud servers, differential privacy processing is performed on multigenerational physiological parameter data. The differential privacy processing includes adding noise that satisfies a Laplace distribution, which is a symmetric exponential distribution centered at zero with a scale parameter determined by the ratio of privacy budget to data sensitivity.

[0110] The privacy budget is obtained by calculating the privacy budget allocation value for each query based on the total privacy budget, the number of queries, and the relaxation parameter, where the relaxation parameter is the upper limit of the small probability deviation allowed in the differential privacy definition.

[0111] The data sensitivity is a value obtained by traversing the dataset and calculating the maximum change in query results caused by the deletion of a single record.

[0112] The environmental physical parameter data and intergenerational activity area data are subjected to coordinate perturbation processing. The coordinate perturbation is a data transformation in which independent and identically distributed Laplace noise is added to the original coordinates.

[0113] Detailed explanation of step S1:

[0114] I. Detailed Implementation Methods for Millimeter-Wave Radar Array Deployment

[0115] The millimeter-wave radar array consists of three 60GHz frequency-modulated continuous wave radar nodes.

[0116] Each node is deployed in a triangular topology on the indoor ceiling.

[0117] The node spacing is 2 to 4 meters to ensure multi-angle coverage of people indoors.

[0118] Each radar node transmits a linear frequency modulated continuous wave with a bandwidth of 4 GHz.

[0119] The frequency sweep period is 1 millisecond.

[0120] Inter-node signal interference is avoided by using time-division multiplexing, which is a time allocation mechanism in which each radar node takes turns transmitting signals according to a preset time slice.

[0121] The depth vision sensor array consists of two time-of-flight depth cameras.

[0122] The installation height is 2.2 meters to 2.5 meters.

[0123] The angle of depression is 15 to 30 degrees.

[0124] The frame rate is 30 frames per second.

[0125] The resolution is 640 pixels × 480 pixels.

[0126] Time synchronization triggering ensures frame synchronization among multiple cameras. This time synchronization triggering is a synchronization mechanism that enables a set number of cameras to expose and acquire data at the same time through a hardware trigger line or a network time protocol.

[0127] II. Detailed Implementation Method for Data Collection on Intergenerational Physiological Coupling Effects

[0128] The heart rate variability sequence was obtained through the following steps:

[0129] The phase change of radar echo caused by micro-movements in the chest cavity is detected by millimeter-wave radar.

[0130] The phase change frequency is used to calculate the heart rate: HR = 60 × fphase, where fphase is the phase change frequency in Hertz.

[0131] Calculate the continuous RR interval: RRn = tpeak^(n+1) - tpeak^(n), where tpeak^(n) is the nth heart rate peak time.

[0132] Calculate heart rate variability: HRVn = RRn - RR(n-1).

[0133] The calculation steps for the Pearson correlation coefficient include:

[0134] Calculate covariance:

[0135]

[0136] Calculate the standard deviation:

[0137]

[0138] Calculate the correlation coefficient:

[0139]

[0140] The respiratory rate sequence was obtained through the following steps:

[0141] The amplitude modulation of radar echoes caused by abdominal undulations is detected by millimeter-wave radar.

[0142] The modulation period is calculated as the breathing period: Tresp = 1 / fmod, where fmod is the amplitude modulation frequency in Hertz.

[0143] Calculate the respiratory rate: RR = 60 / Tresp.

[0144] The instantaneous phase is obtained through Hilbert transform:

[0145] The Hilbert transform of the respiratory rate sequence is: H[RR(t)]=(1 / π)×PV∫{-∞}^{+∞}RR(τ) / (t-τ)dτ, where the Hilbert transform is an integral transform that converts a real-valued signal into an analytic signal.

[0146] Construct the analytic signal: Z(t) = RR(t) + j × H[RR(t)], where the analytic signal is a complex signal with the real part being the original signal and the imaginary part being the result of the Hilbert transform.

[0147] Calculate the instantaneous phase: φ(t)=arg(Zt)=arctan(H[RRt] / RRt), where the instantaneous phase is the argument of the analytic signal in the complex plane.

[0148] The steps for calculating the respiratory synchronization rate include:

[0149] Calculate the phase difference: Δφ(i,j)(t)=φi(t)-φj(t).

[0150] Calculate the complex exponential mean: Rresp^(i,j)=|<exp(j×Δφi,jt)> |=|(1 / T)×Σ(t=1)^Texp(j×Δφi,jt)|, where the average value of the complex exponential function is the average amplitude of the phase difference complex exponential function in the time domain.

[0151] The millimeter-wave radar transmits linear frequency modulated continuous waves in the 60 GHz band.

[0152] The bandwidth is 4GHz.

[0153] The frequency sweep period is 1 millisecond.

[0154] The calculation steps for gait cycle feature extraction include:

[0155] The echo signal is received and mixed with the transmitted signal to obtain an intermediate frequency signal sIF(t), which is a low-frequency signal composed of the frequency difference between the transmitted signal and the echo signal.

[0156] Perform a distance-dimensional Fast Fourier Transform on the intermediate frequency signal to extract the human target distance unit: R=(c×fIF) / (2×K).

[0157] Where c = 3 × 10^8 meters per second is the speed of light.

[0158] fIF is the intermediate frequency, measured in Hertz.

[0159] K = 4 × 10^12 Hz per second is the frequency modulation slope, which is the rate at which the frequency of the transmitted signal changes with time.

[0160] Perform a Doppler Fast Fourier Transform on the intermediate frequency signal within the same distance cell to extract the velocity information of the human target: v=(λ×fD) / 2.

[0161] Where λ = 0.005 meters is the wavelength.

[0162] fD is the Doppler frequency shift, measured in Hertz.

[0163] Temporal analysis of velocity information in consecutive frames is performed to extract gait periodic features: Tgait=1 / fdominant.

[0164] Where fdominant is the dominant frequency component in the power spectrum of the velocity signal, and the dominant frequency component is the frequency corresponding to the peak value of the power spectrum.

[0165] The depth vision sensor acquires indoor depth image sequences at a rate of 30 frames per second.

[0166] The resolution is 640 pixels × 480 pixels.

[0167] The calculation steps for extracting the body contour proportion features include:

[0168] Human detection is performed on depth images to extract human contour bounding boxes. The human detection is an image processing operation that identifies human regions in the image through a deep neural network and outputs bounding rectangles.

[0169] Calculate the height ratio of the human body contour: rheight = Hbody / Havg.

[0170] Where Havg=1.7 meters is the preset average human height.

[0171] Calculate the width-to-height ratio of the human body outline: raspect = Wbody / Hbody.

[0172] Calculate the shoulder-to-hip ratio of the human body contour: shoulderhip = shoulder / hip.

[0173] The height ratio, width-height ratio, and shoulder-hip ratio are concatenated to form the body contour ratio feature vector: fbody=[rheight, raspect, rshoulderhip].

[0174] The calculation steps for the rate of change of activity intensity include:

[0175] Record the position coordinate sequence {(xt,yt)}(t=1)^(Npos) of the i-th generation member within the sampling interval Δt.

[0176] Calculate the position coordinate displacement between adjacent time points: Δdt=sqrt[(xt+1-xt)^2+(yt+1-yt)^2].

[0177] The variance of the position coordinate displacement per unit time is calculated as the rate of change of activity intensity: vi=Var({Δdt}t=1^Npos-1) / Δt.

[0178] The calculation steps for the adaptive sampling frequency adjustment include:

[0179] Statistical analysis of the rate of change of historical activity intensity in the i-th generation {vi^(1),vi^(2),...,vi^(H)}.

[0180] Calculate the mean of the historical activity intensity change rate: μvi=(1 / H)×Σ(h=1)^Hvi^(h).

[0181] The standard deviation of the rate of change of historical activity intensity is calculated as: σvi = sqrt[(1 / H) × Σ(h=1)^H(vi^h-μvi)^2].

[0182] Determine the sampling frequency adjustment condition: If vi > μvi + σvi, then increase the sampling frequency to twice the base frequency f0, i.e., fnew = 2 × f0.

[0183] If vi≤μvi+σvi, then the sampling frequency will be reduced to 0.5 times the fundamental frequency f0, i.e., fnew=0.5×f0.

[0184] The fundamental frequency f0 is the minimum effective frequency required for sampling physiological index data of each generation. It is determined by the Nyquist sampling theorem based on the highest frequency component of the physiological signal. The Nyquist sampling theorem is a signal sampling criterion that the sampling frequency must be greater than twice the highest frequency component of the signal.

[0185] Detailed explanation of environmental regulation implementing agencies:

[0186] A detailed implementation of the environmental control actuator composition includes an air conditioning unit, a fresh air unit, a lighting driver, and a curtain driver.

[0187] The air conditioning unit is a variable frequency multi-split air conditioning indoor unit, which has cooling, heating, dehumidification and air supply modes, and adjusts the cooling output by changing the compressor frequency.

[0188] The fresh air unit is a total heat exchange type fresh air system, which has supply air, exhaust air and bypass modes, and controls the fresh air volume by adjusting the fan speed.

[0189] The lighting driver is a dimmable LED driver power supply with continuous brightness adjustment capability from 0 to 100%, and the output luminous flux is adjusted by changing the PWM duty cycle.

[0190] The curtain driver is an electric curtain track motor with opening, closing, and pause functions. The opening degree of the curtain can be adjusted by changing the rotation angle of the motor.

[0191] A detailed implementation of the environmental control command format is provided, which includes temperature setpoint, fan speed setting, light intensity value, and curtain opening / closing degree.

[0192] The temperature setpoint is the target indoor temperature of the air conditioning unit, with a range of 16 degrees Celsius to 30 degrees Celsius and an accuracy of 0.5 degrees Celsius.

[0193] The fan speed settings refer to the indoor fan speed levels of the air conditioning unit, including low, medium, high, and automatic speeds.

[0194] The luminance value is a percentage of the target output luminous flux of the lighting driver, ranging from 0% to 100%, with an accuracy of 1%.

[0195] The opening degree of the curtain is the percentage of the target opening position of the curtain driver, with a value range of 0% to 100%, where 0% represents fully closed and 100% represents fully open, with an accuracy of 5%.

[0196] The control commands are sent from the control computing unit to each environmental regulation actuator via the MQTT protocol or the Modbus RTU protocol.

[0197] The MQTT protocol is a lightweight message transmission protocol based on the publish / subscribe model, and the ModbusRTU protocol is a master-slave industrial control protocol based on serial communication.

[0198] Regarding the detailed implementation of the measurement of changes and feedback of physical environmental parameters, the indoor temperature is measured by digital temperature sensors deployed in each generational activity area, with a measurement accuracy of ±0.5 degrees Celsius and a sampling period of 1 minute.

[0199] The indoor humidity is measured by capacitive humidity sensors deployed in each generation's activity area, with a measurement accuracy of ±3%RH and a sampling period of 1 minute.

[0200] The light intensity is measured by photodiode illuminance sensors deployed in the activity areas of each generation, with a measurement range of 0 to 2000 lux and a sampling period of 1 minute.

[0201] The air quality index is measured using laser particulate matter sensors and electrochemical gas sensors deployed in various generational activity areas, with a sampling period of 5 minutes.

[0202] The environmental noise level was measured using condenser microphones deployed in each generation's activity area, with a measurement range of 30dB to 90dB and a sampling period of 1 minute.

[0203] The physical environment parameter feedback data refers to the actual physical environment parameter values ​​collected by the aforementioned sensors after the environmental regulation actuator executes the control command.

[0204] The deviation is calculated as the absolute value of the difference between the actual value of the physical environment parameter and the target set value.

[0205] Detailed explanation of step S2:

[0206] Regarding the detailed implementation method for establishing intergenerational physiological baselines, the determination of the 30 consecutive 24-hour cycles is based on the following: experimental verification shows that when the number of historical cycles is less than 20, the baseline statistical stability is insufficient.

[0207] The standard deviation estimation error exceeds 15%.

[0208] When the number of historical periods is 30, the standard deviation estimation error converges to 4.7%.

[0209] When the number of historical periods exceeds 30, the improvement in standard deviation estimation error is less than 1%.

[0210] The experimental data are shown below.

[0211] Historical period count | Distribution mean estimation error | Distribution variance estimation error | Baseline stability score

[0212] 10|8.2%|22.5%|62.3

[0213] 20|4.1%|15.8%|78.6

[0214] 30|2.3%|4.7%|91.2

[0215] 40|1.9%|4.1%|92.5

[0216] 50|1.7%|3.9%|93.1

[0217] Experimental conclusion: The 30 historical periods achieve the optimal balance between estimation error and data accumulation cost.

[0218] Therefore, the number of historical cycles was determined to be 30.

[0219] II. Detailed Implementation of Z-score Offset Normalization

[0220] The calculation steps for Z-score offset normalization include:

[0221] Calculate the distribution mean:

[0222]

[0223] Where Ni = 30 × 1440 = 43200 is the number of samples within 30 days of the i-th generation.

[0224] The sampling frequency is once per minute.

[0225] Calculate the variance of the distribution:

[0226]

[0227] Calculate the Z-score offset:

[0228]

[0229] III. Detailed Implementation Methods for Statistical Analysis of Intergenerational Activity Rhythm Cycles

[0230] The calculation steps for the intergenerational activity rhythm cycle include:

[0231] Record the start and end time series of activities of each generation member within 30 consecutive 24-hour cycles {(tstart^m,tend^m)}(m=1)^30.

[0232] Calculate the activity duration for each cycle: Δt^(m) = tend^(m) - tstart^(m).

[0233] Calculate the intergenerational activity rhythm period: Ti=(1 / 30)×Σ(m=1)^30Δt^(m).

[0234] IV. Detailed Implementation Method for Asynchronous Window Time Alignment

[0235] The calculation steps for asynchronous window alignment include:

[0236] Using Ti as the window length, the data stream xi(t) of the i-th generation is divided into a continuous window sequence.

[0237] wi,l(τ)=xi(ti,l^start+τ), where τ∈[0,Ti], and l is the window index.

[0238] At fusion time t, determine the current window index of each generation: li(t) = floor(t-ti, 0^start / Ti).

[0239] Where ti,0^start is the starting time of the data stream of the i-th generation.

[0240] Calculate the phase offset within the current window for each generation: φi=(t-ti,lt^start) / Ti.

[0241] Among them ti,l(t)^start=ti,0^start+li(t)×Ti.

[0242] Extract the data points corresponding to the phase offset φi for each generation: xi,aligned=wi,li(t)(φi×Ti).

[0243] The data points aligned across generations are used as the fusion input.

[0244] Regarding the detailed implementation of time-frequency feature extraction, the calculation steps for extracting frequency domain features using the short-time Fourier transform include:

[0245] Set the window function to Hanning window: w(n) = 0.5 - 0.5 × cos(2πn / NSTFT-1).

[0246] Where NSTFT is the window length.

[0247] The data stream x(t) is divided into frames with a frame length of NSTFT and a frame shift of NSTFT / 2.

[0248] After windowing each frame of data, perform a Fast Fourier Transform: X(k,f)=Σ(n=0)^(NSTFT-1)x(k×NSTFT / 2+n)×w(n)×exp(-j2πfn / NSTFT).

[0249] The frequency domain feature is extracted as the power spectral density: P(k,f)=|X(k,f)|^2.

[0250] The computational steps for extracting local temporal features by multi-scale wavelet decomposition include:

[0251] The mother wavelet function is selected as the Daubechies fourth-order wavelet, which is a compactly supported orthogonal wavelet basis function with fourth-order vanishing moments.

[0252] Perform a 5-level wavelet decomposition on the data stream: x(t) = A5(t) + Σ(j=1)^5Dj(t).

[0253] Where A5(t) is the fifth-level approximation coefficient and Dj(t) is the j-th-level detail coefficient.

[0254] Extract the energy of each level of detail coefficients as a local feature in the time domain: Ej=Σt|Dj(t)|^2.

[0255] The calculation steps for extracting data distribution features using statistical moments of each order include:

[0256] Calculate the first-order moment (mean): m1=(1 / N)×Σ(n=1)^Nxn.

[0257] Calculate the second-order central moment (variance): m2=(1 / N)×Σ(n=1)^N(xn-m1)^2.

[0258] Calculate the third-order normalized moment (skewness): m3=(1 / N)×Σ(n=1)^N[(xn-m1) / sqrt(m2)]^3.

[0259] Calculate the fourth-order normalized moment (kurtosis): m4=(1 / N)×Σ(n=1)^N[(xn-m1) / sqrt(m2)]^4-3.

[0260] The frequency domain features, local time domain features, and distribution features are concatenated into a feature vector: f=[P(k,f),E1,E2,E3,E4,E5,m1,m2,m3,m4].

[0261] Experimental verification of the determination of the short-time Fourier transform window length: comparative experiments were conducted to determine the optimal window length.

[0262] Experimental setup: Short-time Fourier transform was performed on the same set of multi-generational home environment data using different window lengths.

[0263] Experimental verification of multi-scale wavelet decomposition parameters was conducted, and comparative experiments were performed to determine the optimal mother wavelet and decomposition level.

[0264] Experimental environment conditions: 3-bedroom, 2-living room house, area 120 square meters, room temperature 22±2 degrees Celsius, sample size n=30, number of replicates 3.

[0265] The experimental data are shown below.

[0266] Mother wavelet type | Decomposition level | Temporal local feature discrimination | Computation time (milliseconds) | Reconstruction error

[0267] Haar|3|72.3%|15.2%|8.5%

[0268] Daubechies Level 2 | 4 | 78.6% | 22.5% | 5.2%

[0269] Daubechies Level 4 | 5 | 86.4% | 28.7% | 3.1%

[0270] Daubechies Level 6 | 6 | 84.1% | 35.3% | 2.8%

[0271] Symlets, Level 8 | 5 | 83.7% | 31.8% | 3.5%

[0272] Experimental conclusion: When using the Daubechies fourth-order mother wavelet and the decomposition level is 5, the time-domain local feature discrimination reaches the highest level of 86.4%.

[0273] The reconstruction error was 3.1%.

[0274] The calculation took 28.7 milliseconds.

[0275] Therefore, the mother wavelet is determined to be of Daubechies order 4, and the decomposition level is 5.

[0276] Detailed explanation of step S3:

[0277] A detailed implementation method for measuring intergenerational environmental demand conflict includes the following steps for extracting the intergenerational environmental demand parameter vectors:

[0278] By statistically analyzing historical data, the physiological response of each generation under different environmental parameters was determined, thus identifying the optimal range of environmental parameters for each generation.

[0279] The median of the optimal environmental parameter range is used as the environmental requirement parameter: di^k=(ei,min^k+ei,max^k) / 2.

[0280] Where ei,min^k is the lower limit of the k-th environmental parameter in the i-th generation, and ei,max^k is the upper limit.

[0281] Construct an environment requirement parameter vector:

[0282] di=[di^temp,di^hum,di^light,di^air,di^noise].

[0283] The calculation steps for the cosine distance of the included angle include:

[0284] Calculate the vector inner product: di·dj=Σ(k=1)^5di^k×dj^k.

[0285] Calculate the vector magnitude: ||di||=sqrt[Σ(k=1)^5(di^k)^2], ||dj||=sqrt[Σ(k=1)^5(dj^k)^2].

[0286] Calculate cosine similarity: cos(θi,j)=(di·dj) / (||di||×||dj||).

[0287] Calculate the conflict metric: Ci,j = 1 - cos(θi,j).

[0288] The conflict metric has a value range of [0, 2], and the larger the value, the more severe the conflict.

[0289] Regarding the detailed implementation of the fusion weight allocation, the calculation steps for the fusion weight include:

[0290] Calculate the sum of conflict metrics for each data stream: Ci = Σ(j≠i)Ci,j.

[0291] Calculate the exponent value: exp(Ci)=e^Ci, where e is the base of the natural logarithm.

[0292] Calculate the normalized denominator: Z = Σ(j=1)^Kexp(Cj).

[0293] Calculate the fusion weights:

[0294]

[0295] Verify the weight normalization: Σ(i=1)^Kwi=1.

[0296] Regarding the detailed implementation of multi-level fusion, the calculation steps of the data layer weighted average fusion include:

[0297] The weighted average of the original data streams from each generation is: xfused^raw=Σ(i=1)^Kwi×xi.

[0298] Where wi is the fusion weight of the data stream of the i-th generation, and xi is the original data stream of the i-th generation.

[0299] The weighted average result is used as the data layer fusion output.

[0300] The computational steps for feature layer fusion based on intergenerational attention mechanism include:

[0301] Calculate the intergenerational attention score: ai,j=(fi^TWq)(Wkfj) / sqrt(dk).

[0302] Where fi is the feature vector of the i-th generation, Wq and Wk are the learnable query matrix and key matrix, and dk is the dimension of the key vector.

[0303] The attention score is normalized as follows: αi,j=exp(ai,j) / Σ(m=1)^Kexp(ai,m).

[0304] Calculate the attention-weighted features: ffused=Σ(i=1)^KΣ(j=1)^Kαi,jWvfj.

[0305] Where Wv is the learnable value matrix.

[0306] The attention-weighted features are used as the output of the feature layer fusion.

[0307] The calculation steps for evidence fusion based on conflict metric correction at the decision-making level include:

[0308] Using the decision-making outcomes of each generation as evidence, a basic probability allocation function mi(Ak) is constructed.

[0309] Where Ak is the k-th decision category.

[0310] Calculate the average conflict metric: Ci,avg=(1 / K-1)×Σ(j≠i)Ci,j.

[0311] Calculate the corrected basic probability distribution: mi'(Ak)=mi(Ak)×(1-Ci,avg).

[0312] Normalize the modified basic probability assignment: mi''(Ak)=mi'(Ak) / Σ(l=1)^Lmi'(Al).

[0313] Evidence fusion is performed by applying the combination rule: mfused(Ak)=(1 / 1-N)×Σ(Ap∩Aq=Ak)mp''(Ap)×mq''(Aq).

[0314] Where N=Σ(Ap∩Aq=∅)mp''(Ap)×mq''(Aq) is the conflict coefficient.

[0315] The fused probability allocation is used as the fusion output of the decision layer.

[0316] Detailed explanation of step S4:

[0317] Regarding the detailed implementation of integrating intergenerational work-rest cycle coding, the steps for constructing the sine-cosine periodic function include:

[0318] Determine the periodic parameter: Ti is the activity rhythm period of the i-th generation.

[0319] Construct a sine code: sinenci(t) = sin(2πt / Ti).

[0320] Construct cosine encoding: cosenci(t) = cos(2πt / Ti).

[0321] The sine code and cosine code are concatenated into a periodic code vector: pi(t)=[sinenci(t),cosenci(t)].

[0322] Concatenate the periodic encoding vector with the feature vector sequence: fenhanced(t)=[f(t),pi(t)].

[0323] Where f(t) is the eigenvector at time t.

[0324] The structure of the time-series prediction network includes:

[0325] Input gate: it = σ(Wi × [fenhancedt, ht-1] + bi).

[0326] Where σ is the sigmoid function, Wi is the input gate weight matrix, bi is the input gate bias, and h(t-1) is the hidden state at the previous time step.

[0327] Forget gate: ft = σ(Wf × [fenhancedt, ht-1] + bf).

[0328] Cell state update: Ct=ft⊙C(t-1)+it⊙tanh(WC×[fenhancedt,ht-1]+bC).

[0329] Where ⊙ represents element-wise multiplication.

[0330] Output gate: ot = σ(Wo × [fenhancedt, ht-1] + bo).

[0331] Hidden state update: ht=ot⊙tanh(Ct).

[0332] Output prediction: yt = Wy × ht + by.

[0333] Where yt represents the predicted value of environmental parameters.

[0334] Regarding the detailed implementation of graph-structured neural network construction, the steps for determining the node set include:

[0335] Identify indoor intergenerational activity areas, including bedrooms, living rooms, studies, children's rooms, and rooms for the elderly.

[0336] Each intergenerational activity region is treated as a node in the graph: V={v1,v2,...,vM}.

[0337] The steps for determining the edge set include:

[0338] The physical connections between the inspection areas include ventilation duct connections, door and window openings, and corridor connections.

[0339] If any of the above physical connections exist between region m and region n, then an edge em,n is established between nodes vm and vn.

[0340] The steps for calculating the edge weights include:

[0341] Calculate the overlapping area Am,n: If two regions have spatial overlap, then Am,n is the area of ​​the overlapping region.

[0342] If there is no overlap, then Am,n=0.

[0343] Measure the cross-sectional area Sm,n of the airflow channel: measure the cross-sectional area of ​​the ventilation duct and the opening area of ​​the door and window.

[0344] Calculate edge weights: em,n = Am,n / Sm,n.

[0345] When Am,n=0, em,n=1×10^-6.

[0346] Construct a symmetric adjacency matrix: Am,n=An,m=em,n.

[0347] Regarding the detailed implementation of graph attention networks, the learning steps of the graph attention network include:

[0348] Perform a linear transformation on the node features: hm' = W × hm.

[0349] Where W is a learnable linear transformation matrix.

[0350] Calculate the attention score: em,n=LeakyReLU(a^T[hm'||hn']).

[0351] Where a is the learnable attention vector, and || denotes vector concatenation.

[0352] LeakyReLU(x) = max(0.01x, x).

[0353] The attention score is normalized as follows: αm,n=exp(em,n) / Σ(k∈Nm)exp(em,k).

[0354] Where N(m) is the neighborhood of node m.

[0355] Calculate the attention-weighted output: hm^out=Σ(n∈Nm)αm,n×hn'.

[0356] The environmental impact weight αm,n is used to adjust the predicted values ​​of cross-regional environmental parameters: ym,predict=ym,local+Σ(n∈Nm)αm,n×(yn,local-ym,local).

[0357] Where ym,local represents the local predicted value of node m.

[0358] Detailed explanation of step S5:

[0359] Regarding the detailed implementation of the intergenerational sensitivity matrix construction, the design of the single-factor change experiment includes:

[0360] Determine the range of experimental environmental parameters: temperature 18°C ​​to 28°C, humidity 40%RH to 70%RH, light intensity 200 lux to 800 lux, air quality index 50 to 150, and noise 30 dB to 60 dB.

[0361] Keep the other four environmental parameters at their intermediate values ​​and change only one environmental parameter.

[0362] The variation steps are: temperature ±1°C, humidity ±5%RH, light intensity ±100 lux, air quality index ±10, and noise ±5dB.

[0363] After each change, the body was allowed to stabilize for 10 minutes, and physiological indicators were recorded for each generation.

[0364] Each environmental parameter was tested 10 times repeatedly, and the average value was taken as the data point for the response curve.

[0365] The steps for calculating the response slope include:

[0366] Extract the experimental data of single-factor variation {(e^k,s,xi^k,s)}(s=1)^10.

[0367] Calculate the response slope:

[0368]

[0369] The response slope ki,k represents the sensitivity of the k-th physiological indicator in the i-th generation to changes in the k-th environmental parameter.

[0370] The normalization steps of the intergenerational sensitivity matrix include:

[0371] Calculate the absolute value of the response slope of each index across generations: |ki,k|.

[0372] Calculate the normalized sensitivity weights: wi,k=|ki,k| / Σ(m=1)^KΣ(n=1)^5|km,n|.

[0373] Construct a normalized intergenerational sensitivity matrix: W=[wi,k](K×5).

[0374] II. Detailed Implementation Methods for Constructing Multi-Objective Optimization Functions

[0375] The calculation steps for the intergenerational comfort deviation term include:

[0376] Obtain the actual value of the k-th environmental parameter in the i-th generation: eactual^i,k.

[0377] Obtain the historical preference value epref^i,k of the k-th environmental parameter in the i-th generation.

[0378] The historical preference value is the environmental parameter setting value that appears most frequently in the intergenerational historical environmental adjustment operation record.

[0379] Calculate the squared deviation: (eactual^i,k-epref^i,k)^2.

[0380] Calculate the weighted sum of squares: Jcomfort=Σ(i=1)^KΣ(k=1)^5wi,k×(eactual^i,k-epref^i,k)^2.

[0381] The calculation steps for the system energy consumption item include:

[0382] Obtain the rated power {Pd}(d=1)^D of each environmental control device.

[0383] Record the actual runtime of each device {td}(d=1)^D.

[0384] Calculate the system energy consumption: Jenergy=Σ(d=1)^DPd×td.

[0385] The calculation steps for the device switching frequency item include:

[0386] Set the observation window Twindow to 1 hour.

[0387] Record the number of times each device is started (d=1)^D and the number of times it is turned off (d=1)^D within the observation window.

[0388] Computer device switching frequency:

[0389] Jswitch=Σ(d=1)^D(nd,on+nd,off) / Twindow.

[0390] The steps for constructing the multi-objective optimization function include:

[0391] Determine the target weight coefficients λ1, λ2, and λ3, satisfying λ1 + λ2 + λ3 = 1.

[0392] Construct a multi-objective optimization function:

[0393]

[0394] Regarding the detailed implementation of the policy optimization network solution, the construction of the state space includes: st=[ecurrent,xcurrent,tfeature].

[0395] Where ecurrent is the current environmental parameter vector, xcurrent is the current physiological index vector for each generation, and tfeature is the time feature vector.

[0396] The time feature vector includes time of day, weekday, and season.

[0397] The construction of the action space includes: at=[Δe1,Δe2,...,ΔeD].

[0398] Where Δed is the adjustment amount of the d-th environmental control device.

[0399] The reward function is constructed as follows: Rt = -Jtotal(st,at) - penalty.

[0400] The penalty is a penalty for violating the constraint.

[0401] When the adjustment amount exceeds the device's adjustment range, penalty=1000; otherwise, penalty=0.

[0402] The training steps for the policy optimization network include:

[0403] Initialize the policy generation network μ(s|θ^μ) and the value evaluation network Q(s,a|θ^Q).

[0404] Where θ^μ and θ^Q are network parameters.

[0405] Initialize the target network: θ^μ'←θ^μ, θ^Q'←θ^Q.

[0406] Explore in the simulation environment and store the experience tuple (st, at, Rt, st+1) into the experience replay buffer.

[0407] Randomly sample small batches of experience from the experience replay buffer and update the value assessment network: minimize the loss function L=(1 / N)×Σi(yi-Qsi,ai|θ^Q)^2.

[0408] Among them, yi=Ri+γ×Q'(si+1,μ'si+1|θ^μ')|θ^Q').

[0409] The policy generation network is updated using policy gradient: ∇θ^μJ≈(1 / N)×Σi∇aQ(si,a|θ^Q)|(a=μsi)×∇θ^μμ(si|θ^μ).

[0410] Soft update target network: θ'←τ×θ+(1-τ)×θ'.

[0411] Where τ is the soft update coefficient.

[0412] Repeat the above steps until convergence.

[0413] Detailed explanation of step S6: Detailed implementation method for calculating relative deviation by generation.

[0414] The post-execution data acquisition steps include:

[0415] After the environmental parameter adjustment strategy is executed, wait for the system to stabilize for a time Tsettle.

[0416] The Tsettle duration is 5 minutes.

[0417] After stabilization, multigenerational physiological index data {xi,after^k}(i=1,k=1)^(K,4) and environmental parameter data {eafter^k}(k=1)^5 were collected.

[0418] The steps for calculating the change include:

[0419] Calculate the change in the k-th physiological indicator in the i-th generation: Δxi^k = xi,after^k - xi,before^k.

[0420] The change refers to the amount of change in physiological indicators after the implementation of the regulation strategy.

[0421] The relative deviation calculation steps include:

[0422] Obtain the expected target value xi,target^k of the k-th physiological indicator in the i-th generation.

[0423] The expected target value is the central value of the intergenerational physiological baseline, μi^k.

[0424] Calculate the relative deviation: δi^k=Δxi^k / xi,target^k.

[0425] The relative deviation represents the proportion of the change relative to the expected target.

[0426] The steps for calculating the comprehensive relative deviation include:

[0427] Calculate the square of the relative deviation of each indicator: (δi^k)^2.

[0428] Calculate the overall relative deviation: δi=sqrt[Σ(k=1)^4(δi^k)^2] / 4.

[0429] The overall relative deviation is the root mean square value of the relative deviation of each indicator.

[0430] Regarding the detailed implementation of the re-decision condition judgment, the historical relative deviation statistics step includes:

[0431] Collect the relative deviation data of the i-th generation in the historical H adjustments {δi^(1),δi^(2),...,δi^(H)}.

[0432] The historical relative deviation data is stored in a historical database.

[0433] The steps for calculating the statistical mean include:

[0434] The arithmetic mean of the historical relative deviation is calculated as: μδi=(1 / H)×Σ(h=1)^Hδi^(h).

[0435] The statistical mean represents the average level of historical adjustment effect.

[0436] The steps for calculating the statistical standard deviation include:

[0437] Calculate the squared deviation of each historical relative deviation from the mean: (δi^h-μδi)^2.

[0438] Calculate the arithmetic mean of the squared deviations: Varδi=(1 / H)×Σ(h=1)^H(δi^h-μδi)^2.

[0439] Calculate the statistical standard deviation: σδi=sqrt(Varδi).

[0440] The steps for determining the conditions for intergenerational re-decision include:

[0441] Calculate the current overall relative deviation δi.

[0442] Judgment condition: If δi>2×σδi, then trigger the return step S1.

[0443] The condition indicates that the current adjustment effect deviates from the historical average by more than 2 standard deviations, and is therefore judged as an adjustment anomaly.

[0444] The global weighted bias calculation steps include:

[0445] Obtain the priority weights {wi}(i=1)^K for each generation.

[0446] Calculate the global weighted bias: δglobal=Σ(i=1)^Kwi×δi.

[0447] The global weighted bias is the weighted sum of the biases of each generation.

[0448] The steps for calculating the global historical weighted bias statistical standard deviation include:

[0449] Collect historical global weighted bias data {δglobal^(1),δglobal^(2),...,δglobal^(H)}.

[0450] Calculate the historical global weighted deviation statistical mean: μδglobal=(1 / H)×Σ(h=1)^Hδglobal^(h).

[0451] Calculate the historical global weighted bias statistical standard deviation: σδglobal=sqrt[(1 / H)×Σ(h=1)^H(δglobal^h-μδglobal)^2].

[0452] The global re-decision condition judgment step includes:

[0453] Calculate the current global weighted bias δglobal.

[0454] Judgment condition: If δglobal > 1.5 × σδglobal, then trigger the return step S1.

[0455] The condition indicates that the global adjustment effect deviates from the historical average by more than 1.5 standard deviations, and a new decision is required.

[0456] Regarding the experimental verification of the determination of the re-decision threshold, the following experiments were conducted to determine the specific value of the preset multiple:

[0457] Experimental setup: In historical adjustment data, different multiple thresholds were used as re-decision conditions, and the false trigger rate and missed trigger rate were statistically analyzed.

[0458] Detailed explanation of the environmental preference model:

[0459] Regarding the detailed implementation of the environmental preference model, the determination of the continuous 3-month data accumulation period is based on the following: experimental verification shows that when the data accumulation period is less than 1 month, the preference distribution statistics are unstable.

[0460] The expected value estimation error exceeds 18.5%.

[0461] When the data accumulation period is 3 months, the expected value estimation error converges to 4.7%.

[0462] When the data accumulation period exceeds 3 months, the improvement in the expected value estimation error is less than 0.6%.

[0463] Regarding the detailed implementation of the robust adjustment strategy generation, the calculation steps for the 95% confidence interval include:

[0464] Calculate the standard error: SEi,s,k=sqrt(Vari,s,k / Ni,s,k).

[0465] Determine the statistical critical value t0.025,Ni,s,k-1 corresponding to a confidence level of 95%.

[0466] Where the degrees of freedom df = Ni,s,k-1.

[0467] Calculate the lower limit of the confidence interval: CIlower=Ei,s,k-t0.025,Ni,s,k-1×SEi,s,k.

[0468] Calculate the upper limit of the confidence interval: CIupper=Ei,s,k+t0.025,Ni,s,k-1×SEi,s,k.

[0469] The confidence interval [CIlower, CIupper] represents the 95% confidence range of the generation's preference for this environmental parameter in this scenario.

[0470] The steps for generating the robust adjustment strategy include:

[0471] Construct a robust optimization objective: minemaxepref∈CIJcomfort(e,epref).

[0472] Where CI is the confidence interval and Jcomfort is the intergenerational comfort deviation term.

[0473] Transform the robust optimization objective into an equivalent form: mine, tt.

[0474] The constraint is Jcomfort(e,epref)≤t for all epref∈CI.

[0475] Since the confidence interval is a continuous interval, a sampling approximation method is adopted: 100 points {epref^(m)} (m=1)^100 are uniformly sampled within the confidence interval.

[0476] The constraint is approximated as a finite number: Jcomfort(e,epref^m)≤t, m=1,2,...,100.

[0477] Solving the above approximate optimization problem yields the robust adjustment strategy erobust.

[0478] The robust adjustment strategy minimizes the intergenerational comfort bias term within the 95% confidence interval of the preference distribution.

[0479] Detailed explanation of edge-cloud collaboration and differential privacy:

[0480] For a detailed implementation of the local-remote hierarchical deployment, the local edge computing node is the NVIDIA Jetson AGXXavier embedded computing platform.

[0481] It features a 512-core Volta graphics processing unit.

[0482] The computing power is 32 TOPS.

[0483] The power consumption is 30W.

[0484] The local edge computing node includes an ARM processor, a graphics processing acceleration unit, and a 32GB memory storage unit.

[0485] The remote cloud server is a cloud server instance equipped with an NVIDIA A100 graphics processing unit.

[0486] The local deployment tasks include: data cleaning (outlier removal, missing value imputation), denoising (Kalman filtering, wavelet threshold denoising), feature extraction (short-time Fourier transform, wavelet decomposition), and data layer weighted average fusion.

[0487] The remote deployment tasks include: temporal prediction using a temporal prediction network, spatial correlation analysis using a graph-structured neural network, solving multi-objective optimization functions, and training a policy optimization network.

[0488] The detailed calculation steps for privacy budget allocation include:

[0489] Set the total privacy budget to εtotal=5.0.

[0490] The relaxation parameter is determined to be δ = 1 × 10^-5, which is the upper limit of the small probability deviation allowed in the differential privacy definition.

[0491] Set the number of queries to T=100 (number of queries per day).

[0492] The quadratic equation is solved iteratively: εtotal=sqrt(2T×ln1 / δ)×εper+T×εper^2 / 2.

[0493] Substituting the values: 5.0 = sqrt(200 × ln10^5) × εper + 50 × εper^2 = 24.0 × εper + 50 × εper^2.

[0494] Solve the quadratic equation: 50εper^2 + 24εper - 5 = 0.

[0495]

[0496] Therefore, the privacy budget for each query is set at εper = 0.157.

[0497] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for data fusion, perception, and optimization of smart home environments, characterized in that, Includes the following steps: S1. Collect multi-generational physiological parameter data, environmental physical parameter data, intergenerational activity area data, and intergenerational physiological coupling effect data through the distributed sensor array; The distributed sensing array includes a millimeter-wave radar array and a depth vision sensor array. The millimeter-wave radar array is a distributed sensing array composed of a set number of 60GHz frequency-modulated continuous wave radar nodes. The depth vision sensor array is a distributed vision array composed of a set number of time-of-flight depth cameras. The multigenerational physiological parameter data includes non-contact heart rate, respiratory rate, and body surface temperature. The non-contact heart rate is the heart rate value obtained by converting the radar echo phase change frequency caused by micro-movement of the chest cavity by millimeter-wave radar. The respiratory rate is the respiratory rate value obtained by converting the radar echo amplitude modulation period caused by abdominal undulation by millimeter-wave radar. The body surface temperature is the temperature value obtained by collecting the temperature distribution of the facial area by an infrared thermal imaging sensor and calculating the regional weighted average. The environmental physical parameter data include indoor temperature, indoor humidity, light intensity, air quality index, and environmental noise level; The intergenerational activity area data consists of the location coordinates and dwell time of each generation member in the indoor space. The location coordinates are obtained by mapping the center point of the human body contour collected by the depth vision sensor to the indoor plane coordinate system. The intergenerational physiological coupling effect data are the correlation coefficient of heart rate variability and respiratory synchronization rate among members of different generations in the same space. S2. Based on the statistical analysis of multi-generational physiological parameter data within a preset historical period, establish a benchmark map of intergenerational physiological characteristics. Normalize the current multi-generational physiological parameter data relative to the benchmark map of intergenerational physiological characteristics by offset. Asynchronously align the multi-source data streams with the intergenerational activity rhythm period as the window length, and extract time-frequency features to construct a feature vector sequence. The intergenerational physiological characteristic baseline map is a statistical baseline interval characterizing the normal fluctuation range of various physiological parameters of each generation member, and is used as the physiological basis for judging environmental regulation needs. The intergenerational activity rhythm cycle is the arithmetic mean of the start and end time intervals of activities of each generation member within a preset historical cycle. S3. Calculate the intergenerational environmental demand conflict measure based on the intergenerational environmental demand parameter vectors, allocate fusion weights according to the conflict measure, and perform multi-level fusion of feature vector sequences in the data layer, feature layer and decision layer to generate comprehensive environmental status assessment results. Use the comprehensive environmental status assessment results to predict the changing trend of environmental physical parameters through a time series prediction network, and model the heat and mass transfer correlation between different intergenerational activity areas through a graph structure neural network. The conflict metric is the cosine distance between the environmental demand parameter vectors of different generations, and the environmental demand parameter vector is a multidimensional vector composed of the optimal demand values ​​of each generation for multiple environmental physical parameters. The heat and mass transfer correlation refers to the intensity of physical transfer of heat, humidity and airflow between different generational activity areas through the building ventilation structure. S4. Based on the changing trends of environmental physical parameters and the correlation of heat and mass transfer, a multi-objective optimization function is constructed by combining the intergenerational sensitivity matrix. An environmental regulation control command is generated by solving the strategy optimization network and sent to the environmental regulation actuator. The environmental regulation actuator changes the indoor physical environmental parameters according to the environmental regulation control command. The environmental control actuators include air conditioning units, fresh air units, lighting actuators, and curtain actuators; The environmental regulation control commands include temperature setpoint, fan speed setting, light intensity value, and curtain opening / closing degree; The multi-objective optimization function includes an intergenerational comfort deviation term, a system energy consumption term, and an equipment switching frequency term. The intergenerational comfort deviation term is the weighted sum of squares of the differences between the actual environmental physical parameters of each generation and the corresponding historical preferred environmental physical parameters of the generation. The system energy consumption term is the total power consumption of the current environmental regulation actuators. The equipment switching frequency term is the number of times the environmental regulation actuators start and stop per unit time. The generational sensitivity matrix is ​​obtained by statistically analyzing the response slope of historical physiological parameter data of each generation to changes in environmental physical parameters. The response slope is the linear fitting coefficient between the changes in physiological parameters of each generation and the changes in environmental physical parameters. S5. Collect feedback data of indoor physical environment parameters and multi-generational physiological parameters after execution, calculate the deviation between the actual value of physical environment parameters and the target set value. When the deviation of any generation exceeds the preset multiple of the statistical standard deviation of the historical deviation of that generation, or when the weighted sum of the deviations of a set number of generations exceeds the preset multiple of the statistical standard deviation of the global historical weighted deviation, return to step S1.

2. The method for data fusion, perception, and optimization of smart home environments according to claim 1, characterized in that, The collection and use of intergenerational physiological coupling effect data in step S1 includes: Simultaneously collect heart rate variability and respiratory rate sequences of different generations of members in the same enclosed indoor space using a millimeter-wave radar array. The heart rate variability sequence is a time series data composed of the interval differences between consecutive heartbeats. The heart rate is obtained by converting the frequency of radar echo phase changes caused by micro-movements in the chest cavity by millimeter-wave radar. The interval differences between consecutive heartbeats are then calculated. The respiratory rate sequence is a time series data composed of respiratory cycles, which is obtained by converting the radar echo amplitude modulation period caused by abdominal undulation detected by millimeter-wave radar. Calculate the correlation coefficient of heart rate variability: Calculate the ratio of the product of the covariance and the standard deviation of the heart rate variability sequences of the i-th generation member and the j-th generation member to obtain the correlation coefficient of heart rate variability, which characterizes the degree of synchronization of heart rate fluctuations between the two generations. Calculate the respiratory synchronization rate: Perform phase synchronization analysis on the respiratory rate sequences of the i-th generation member and the j-th generation member to obtain the respiratory synchronization rate, which characterizes the degree of synchronization of the respiratory rhythms of the two generations. The phase synchronization analysis is a signal processing operation that calculates the complex exponential average of the phase difference after obtaining the instantaneous phase by performing Hilbert transform on the respiratory rate sequence. In step S3, the weight allocation of each generation region in the environmental status assessment is adjusted according to the intergenerational physiological coupling effect data. The adjustment is as follows: when the heart rate variability correlation coefficient is greater than the first preset threshold and the respiratory synchronization rate is greater than the second preset threshold, the two generation members are determined to be in a strong coupling state, and the weight allocation ratio of the strong coupling state generation region in the environmental status assessment is increased to a preset multiple of the original weight allocation ratio. The original weight allocation ratio is the weight allocation ratio value allocated in step S3 through conflict metric sum exponential calculation and normalization. The first preset threshold and the second preset threshold are determined through experimental verification. The experimental verification involves collecting heart rate variability correlation coefficient and respiratory synchronization rate data of multiple intergenerational combinations in a controlled environment, and selecting the threshold parameters after statistically analyzing the boundary values ​​between strong coupling and weak coupling states.

3. The method for data fusion, perception, and optimization of smart home environments according to claim 2, characterized in that, The determination of intergenerational membership and adaptive sampling in step S1 includes: The millimeter-wave radar array transmits a frequency-modulated continuous wave signal in the 60GHz band, and after receiving the echo signal, it performs frequency mixing to obtain an intermediate frequency signal. The intermediate frequency signal is a low-frequency signal composed of the frequency difference between the transmitted signal and the echo signal. The intermediate frequency signal is subjected to distance-dimensional fast Fourier transform and Doppler-dimensional fast Fourier transform to extract the Doppler frequency shift information of the human target as gait periodic features. The gait periodic features are the reciprocals of the dominant frequency components of the power spectrum of the velocity signal during human walking. Indoor depth image sequences are acquired at a rate of 30 frames per second using a depth vision sensor array. The height ratio, width-to-height ratio, and shoulder-to-hip ratio of the human body contour are extracted as body contour proportion features. The height ratio is the ratio of the height of the human body contour to the preset average human body height. The width-to-height ratio is the ratio of the width of the human body contour to the height. The shoulder-to-hip ratio is the ratio of the width of the shoulder to the width of the hip of the human body contour. Gait cycle features and body shape proportion features are concatenated to form an intergenerational biological feature vector, which is input into a classification model based on radial basis function kernel function, and outputs intergenerational identity as elderly member, adult member or minor member. The radial basis function kernel function is a kernel function that takes the Euclidean distance between feature vectors as input and outputs a similarity value through exponential decay operation. The classification model determines the regularization parameters and kernel function parameters through parameter optimization experiments. The parameter optimization experiments are an experimental process of evaluating the model accuracy and selecting the optimal parameter combination within a preset parameter range through cross-validation. The sampling frequency of the sensor nodes corresponding to each generation's activity area in the distributed sensing array is adaptively adjusted according to the rate of change of activity intensity of each generation member. The rate of change of activity intensity is determined by the variance of the displacement of position coordinates per unit time, and the displacement of position coordinates is the Euclidean distance between position coordinates at adjacent sampling times. When the rate of change in activity intensity is greater than the sum of the mean and standard deviation of the historical rate of change in activity intensity for that generation, the acquisition frequency of the corresponding sensor nodes is increased to twice the base frequency. When the rate of change of activity intensity is less than or equal to the sum of the mean and standard deviation of the historical rate of change of activity intensity in that generation, the acquisition frequency of the corresponding sensor nodes in the region is reduced to 0.5 times the base frequency. The fundamental frequency is the lowest effective frequency required for sampling physiological parameter data across generations, determined by the Nyquist sampling theorem based on the highest frequency component of the physiological signal.

4. The smart home environment data fusion sensing and optimization method according to claim 3, characterized in that, The specific implementation of establishing the intergenerational physiological characteristic baseline map and extracting time-frequency features in step S2 includes: The physiological parameter data of each generation member are statistically analyzed within a continuous preset historical period. The arithmetic mean of each physiological parameter data of each generation is calculated as the distribution mean. The square root of the arithmetic mean of the squares of the deviations of each physiological parameter data of each generation from the distribution mean is calculated as the distribution variance. The distribution mean is used as the baseline center and the distribution variance of the preset multiple is used as the baseline fluctuation range to establish a benchmark map of intergenerational physiological characteristics. The intergenerational physiological characteristic baseline map is a statistical baseline interval characterizing the normal fluctuation range of various physiological parameters of each generation member, and is used as the physiological basis for judging environmental regulation needs. The current multigenerational physiological parameter data is normalized relative to the corresponding generational physiological characteristic baseline map by offset. The offset normalization is a data transformation process that subtracts the center value of the corresponding generational physiological characteristic baseline map from the current physiological parameter data and then divides it by half the width of the fluctuation range of the corresponding generational physiological characteristic baseline map. The intergenerational activity rhythm cycle is obtained by statistically analyzing the historical activity records of each generation member. The intergenerational activity rhythm cycle is the arithmetic mean of the start and end time intervals of the activities of each generation member within a continuous preset historical cycle. The start and end times of the activities are the start and end times of the continuous period in which the change in the position coordinates of indoor personnel exceeds a preset displacement threshold. Asynchronous window alignment is performed on multi-source data streams using the intergenerational activity rhythm cycle as the window length, and time-frequency features are extracted to construct a feature vector sequence. The time-frequency features are used to identify environmental disturbance features corresponding to different intergenerational activity patterns. The time-frequency features include: The frequency domain features are extracted by short-time Fourier transform, which uses a Hanning window with a window length of one-quarter of the intergenerational activity rhythm period and a frame shift of one-half of the window length. The time-domain local features are extracted through multi-scale wavelet decomposition, which uses compactly supported orthogonal wavelet basis functions with fourth-order vanishing moments and has a decomposition level of 5. The data distribution characteristics extracted by statistical moments of various orders include the first-order raw moment, the second-order central moment, the third-order standardized moment, and the fourth-order standardized moment. The first-order raw moment is the arithmetic mean of the data, the second-order central moment is the arithmetic mean of the squared deviations of the data from the mean, the third-order standardized moment is the third moment after data standardization, and the fourth-order standardized moment is the fourth moment after data standardization minus 3.

5. The method for data fusion, perception, and optimization of smart home environments according to claim 4, characterized in that, The specific implementation of multi-level fusion in step S3 includes: The data layer adopts a weighted average fusion based on fusion weights. The weighted average fusion is the sum of the products of each generation's original data stream and the corresponding fusion weight, which is used to eliminate the dimensional differences between different generations of data streams. The feature layer employs a fusion based on an intergenerational attention mechanism, which calculates intergenerational attention scores and performs normalization to obtain attention-weighted features, which are used to extract correlation features between different intergenerational data streams. The decision-making level employs evidence fusion based on conflict metric correction, which includes adjusting the basic probability allocation values ​​of each generation's evidence according to the conflict metric, in order to resolve conflicts between decision outcomes of different generations. The progressive logic of the multi-level fusion is as follows: after the data layer fusion eliminates the difference in dimensions, it is input to the feature layer fusion to extract related features; after the feature layer fusion outputs the related features, it is input to the decision layer fusion to resolve decision conflicts; and the three-level fusion outputs the comprehensive environmental state assessment result, which is used to generate subsequent environmental regulation and control instructions.

6. The smart home environment data fusion sensing and optimization method according to claim 5, characterized in that, The edge weights of the graph structure neural network described in step S3 are determined by the ratio of the overlapping area of ​​the intergenerational activity regions to the cross-sectional area of ​​the airflow channels between the connected regions. The overlapping area is the planar area of ​​the part where the two intergenerational activity areas intersect on the indoor space plan. The cross-sectional area of ​​the airflow channel is obtained by measuring the following steps: For ventilation ducts, measure the inner diameter of the duct and calculate the cross-sectional area, which is the product of pi and the square of half the inner diameter of the duct. For door and window openings, measure the width and height of the opening and calculate the cross-sectional area, which is the product of the opening width and the opening height. For regions without physical connections, the edge weights are set to a minimum positive number, which is the lower limit of the machine's floating-point precision. The edge weights are used to model the intensity of physical transfer of heat, humidity, and airflow through the building's ventilation structure between different generational activity areas.

7. The method for data fusion sensing and optimization of smart home environment according to claim 6, characterized in that, The construction of the intergenerational sensitivity matrix in step S4 includes: The other four environmental physical parameters are fixed at the median value. Only one environmental physical parameter is changed, and the corresponding physiological parameter change is recorded. The median value is the arithmetic mean of the preset effective range of the environmental physical parameters. The response slope of each generation of physiological parameter data to changes in environmental physical parameters is calculated. The response slope is the linear fitting coefficient of the changes in physiological parameters and environmental physical parameters in each generation. The linear fitting coefficient is the regression coefficient obtained by fitting the linear relationship between the changes in physiological parameters and environmental physical parameters. After normalizing the response slopes, an intergenerational sensitivity matrix is ​​constructed. The normalization is an operation in which the absolute value of each response slope is divided by the sum of the absolute values ​​of all response slopes. The intergenerational sensitivity matrix is ​​used to quantify the physiological sensitivity of different generations to temperature, humidity, light and noise, and serves as the basis for the weighting of the intergenerational comfort deviation term in the multi-objective optimization function.

8. The method for data fusion, perception, and optimization of a smart home environment according to claim 7, characterized in that, The solution to the multi-objective optimization function in step S5 adopts a strategy optimization network based on dynamic adjustment of generational priorities. The policy optimization network is a deep neural network structure that includes a policy generation network and a value evaluation network. The policy generation network is a deep neural network that includes an input layer, a hidden layer, and an output layer. The hidden layer uses a non-linear activation function, and the output layer uses a bounded output activation function to generate control commands for air conditioning units, fresh air units, lighting drivers, and curtain drivers. The value evaluation network is a deep neural network containing an input layer, a hidden layer, and an output layer, used to evaluate the expected returns of the adjustment strategy. The intergenerational priority is dynamically updated based on the satisfaction statistics of historical adjustment feedback of each generation. The satisfaction is the ratio of the absolute value of the Z-score offset of the physiological index after adjustment to the absolute value of the Z-score offset of the physiological index before adjustment. When the absolute value of the Z-score offset before adjustment is zero, the satisfaction is the absolute value of the Z-score offset after adjustment.

9. The method for data fusion sensing and optimization of smart home environment according to claim 1, characterized in that, include: Establish an intergenerational environmental preference model. By accumulating data over a continuously preset period, the distribution of environmental physical parameter preferences of different generations in different activity scenarios is statistically analyzed. The activity scenarios include sleep scenarios, dining scenarios, learning scenarios, leisure scenarios, and sports scenarios. The sleep scenario is the time spent resting in bed at night. The dining scenario is the time spent eating in a restaurant or kitchen. The learning scenario is the time spent reading and writing in a study or at a desk. The leisure scenario is the time spent watching and listening to entertainment in the living room. The sports scenario is the time spent engaging in physical activities indoors. Calculate the expected value and variance of the environmental physical parameter preference distribution, where the expected value is the arithmetic mean of historical environmental physical parameter setting data, and the variance is the arithmetic mean of the squared deviations between the historical environmental physical parameter setting data and the expected value. In step S4, a robust environmental setting value is generated by combining the expected value and variance of the environmental physical parameter preference distribution. The robust environmental setting value is the target setting value of the environmental physical parameters that minimizes the intergenerational comfort deviation term within a preset confidence interval. The lower limit of the preset confidence interval is the expected value minus the product of the statistical critical value and the standard error. The upper limit of the preset confidence interval is the expected value plus the product of the statistical critical value and the standard error. The statistical critical value is the t-distribution critical value obtained by looking up the table based on the sample degrees of freedom. The standard error is the square root of the quotient of the variance and the sample size.

10. The method for data fusion, perception, and optimization of a smart home environment according to claim 1, characterized in that, Also includes: The real-time processing of physiological parameter data involving multiple generations is deployed on a local edge computing node. The local edge computing node is an embedded computing platform that includes an ARM processor, a graphics processing acceleration unit, and a memory storage unit of no less than 32GB. The local edge computing node is directly connected to the distributed sensor array and the environmental regulation actuator through a wired or wireless communication interface. The environment model reasoning and global optimization solution are deployed on a remote cloud server. Before transmitting data between local edge computing nodes and remote cloud servers, differential privacy processing is performed on multigenerational physiological parameter data. The differential privacy processing includes adding noise that satisfies a Laplace distribution, which is a symmetric exponential distribution centered at zero with a scale parameter determined by the ratio of privacy budget to data sensitivity. The privacy budget is obtained by calculating the privacy budget allocation value for each query based on the total privacy budget, the number of queries, and the relaxation parameter, where the relaxation parameter is the upper limit of the small probability deviation allowed in the differential privacy definition. The data sensitivity is a value obtained by traversing the dataset and calculating the maximum change in query results caused by the deletion of a single record. The environmental physical parameter data and intergenerational activity area data are subjected to coordinate perturbation processing. The coordinate perturbation is a data transformation in which independent and identically distributed Laplace noise is added to the original coordinates.

Images