Method for personalized regulation of indoor environment fusing human physiological parameters

Abstract

The application relates to the technical field of indoor environment regulation, and discloses an indoor environment personalized regulation method fusing human physiological parameters, which acquires the heart rate, the respiration rate and the body surface temperature distribution of a user through a millimeter wave radar and infrared thermal imaging, establishes an individual physiological benchmark library and constructs a personalized utility function; and introduces a Fanger thermal comfort equation for constraint through a physical information neural network; triggers feedforward control when there is a risk of decline in the thermal comfort index, determines a target control parameter adjustment instruction in combination with Bayesian optimization and executes; and continuously updates the physiological parameters and the utility function, thereby forming a closed-loop personalized regulation process. The application extracts the heart rate, the respiration rate and the body surface temperature distribution by fusing millimeter wave radar data and infrared thermal imaging data, and predicts the thermal comfort index in combination with environmental parameters, so that the indoor environment regulation no longer depends on environmental sensor feedback, thereby solving the problem that the human physiological state is neglected in the prior art.

Description

Technical Field

[0001] This invention relates to the field of indoor environment control technology, and in particular to a personalized indoor environment control method that integrates human physiological parameters. Background Technology

[0002] With the development of indoor environmental control technology, air conditioning systems have been widely used in residential, office, and public spaces to regulate air temperature, humidity, and airflow. Existing indoor environmental control systems typically acquire environmental parameters through devices such as temperature sensors, humidity sensors, and wind speed sensors, and adjust the operating status of air conditioning equipment based on these collected parameters to maintain the indoor environment within a preset range.

[0003] Current technologies typically regulate the indoor environment by detecting the deviation between the indoor air temperature and the set temperature, and combining this with parameters such as humidity and wind speed. The controller then adjusts the air conditioning unit's supply air temperature, speed, and volume to achieve this. However, due to differences in age, gender, clothing, and activity level, individuals experience significant variations in their actual thermal sensations within the same indoor environment. Environmental parameters alone cannot accurately reflect an individual's true thermal comfort. Furthermore, existing control mechanisms based on environmental sensor feedback are passive and lag-dependent; by the time the sensors detect a deviation from the set value, discomfort has often already developed.

[0004] However, the inventors of this application discovered in the process of realizing the technical solution of this application that the above-mentioned technology has at least the following technical problems: the existing indoor environmental control system takes environmental parameters as the main basis for adjustment, lacks direct perception and modeling of human physiological state, and the control process has response lag, resulting in a deviation between the environmental adjustment result and the real-time physiological needs of the human body, and cannot achieve truly personalized comfort control. Summary of the Invention

[0005] To overcome the above shortcomings, this invention provides a personalized indoor environment control method that integrates human physiological parameters, aiming to improve the problem that the existing technology lacks direct perception and modeling of human physiological state, resulting in the inability to reflect the actual thermal comfort state of the human body during the control process.

[0006] This invention provides the following technical solution: a method for personalized indoor environment control that integrates human physiological parameters, comprising the following steps:

[0007] S1. Acquire millimeter-wave radar data and infrared thermal imaging data of indoor users, and extract heart rate and respiratory rate based on the millimeter-wave radar data, and extract body surface temperature distribution based on the infrared thermal imaging data.

[0008] S2. Based on the heart rate, respiratory rate and body surface temperature distribution, establish an individual physiological benchmark library corresponding to the user, and construct a personalized utility function based on the individual physiological benchmark library. The personalized utility function is a decision model used to characterize the user's preference for different environmental states and different air conditioning equipment operating parameters.

[0009] S3. Input the physiological parameter data, environmental parameter data and air conditioning equipment operation parameters of the current time and historical time into the time series prediction network to predict the trend of physiological parameter change and thermal comfort index change in the future time. The thermal comfort index is a comprehensive thermal comfort evaluation index calculated based on the physiological parameter data and the environmental parameter data.

[0010] S4. The prediction process of the time series prediction network is constrained by a physical information neural network, wherein the physical information neural network uses the Fanger thermal comfort equation as a physical constraint.

[0011] S5. When the predicted thermal comfort index is at risk of falling to a preset threshold, control parameter adjustment instructions for the air conditioning equipment are generated according to the personalized utility function to achieve feedforward control.

[0012] S6. Optimize the control parameter adjustment command through Bayesian optimization, determine the target control parameter adjustment command under the premise of satisfying comfort constraints, and control the air conditioning equipment to execute the target control parameter adjustment command.

[0013] Preferably, in step S1, the step of extracting heart rate and respiratory rate based on the millimeter-wave radar data includes:

[0014] The millimeter-wave radar data is subjected to target range gating to determine the main reflection area corresponding to the indoor user;

[0015] Phase unwrapping and detrending processing are performed on the echo signal corresponding to the main reflection region to obtain a phase sequence characterizing the micro-motion changes of the thoracic cavity;

[0016] The phase sequence was subjected to Doppler filtering and frequency band separation to obtain the respiratory signal component and the heartbeat signal component, respectively.

[0017] The respiratory signal component and the heartbeat signal component are subjected to spectral peak detection to determine the respiratory rate and the heart rate, respectively.

[0018] Preferably, in step S1, the step of extracting the body surface temperature distribution based on the infrared thermal imaging data includes:

[0019] The infrared thermal imaging data is segmented into human body regions to obtain the body surface regions corresponding to indoor users;

[0020] Based on the aforementioned body surface region, the average body surface temperature and body surface temperature difference features are extracted.

[0021] The average body surface temperature and the body surface temperature difference characteristics are used as the characterization results of the body surface temperature distribution;

[0022] The body surface temperature distribution is time-aligned with the heart rate and respiratory rate to form physiological parameter data corresponding to the same moment.

[0023] Preferably, in step S2, the step of establishing an individual physiological baseline library corresponding to the user based on the heart rate, the respiratory rate, and the body surface temperature distribution includes:

[0024] The heart rate, respiratory rate and body surface temperature distribution of the user were collected at multiple time periods to form an individual baseline sample set;

[0025] The individual physiological benchmark library is established based on the individual benchmark sample set to characterize the user's physiological characteristics and fluctuation range in a resting state;

[0026] The personalized utility function is constructed based on the individual physiological benchmark library and the user's historical physiological parameter change trends.

[0027] The personalized utility function is used to evaluate preferences for different environmental conditions and different combinations of air conditioning equipment operating parameters.

[0028] Preferably, in step S2, the step of constructing the personalized utility function includes:

[0029] Establish a set of candidate control strategies, with different candidate control strategies corresponding to different combinations of air conditioning equipment operating parameters;

[0030] Reward information is determined based on the user's feedback or behavioral changes after executing the candidate control strategy;

[0031] The reward information is learned based on the multi-armed slot machine algorithm to determine the preference strategy that matches the user;

[0032] The personalized utility function is updated according to the preference strategy.

[0033] Preferably, in step S3, the steps of predicting the future trends of physiological parameters and thermal comfort index include:

[0034] Construct a time series input sample consisting of physiological parameter data, environmental parameter data, and air conditioning equipment operating parameters from multiple consecutive time points;

[0035] The time-series input samples are then fed into the LSTM-Attention network.

[0036] Extracting temporal dependency features from time series using an LSTM network;

[0037] The features at different historical moments are weighted using the Attention mechanism;

[0038] Based on the weighted temporal features, the output shows the future trends of physiological parameter changes and thermal comfort index changes.

[0039] Preferably, in step S4, the step of constraining the prediction process of the time-series prediction network using a physical information neural network includes:

[0040] Obtain the thermal comfort index prediction results output by the time-series prediction network;

[0041] Based on the physiological and environmental parameter data, the physical constraint results of the thermal comfort index are calculated using the Fanger thermal comfort equation.

[0042] The model training process is corrected based on the deviation between the predicted thermal comfort index and the physical constraint results of the thermal comfort index.

[0043] The training of the physical information neural network is completed by jointly optimizing the data fitting term and the physical constraint term.

[0044] Preferably, in step S5, the step of generating control parameter adjustment instructions for the air conditioning equipment based on the personalized utility function includes:

[0045] Based on the future trend of thermal comfort index changes, determine whether there is a risk of it dropping to a preset threshold;

[0046] In the event of the aforementioned risk, feedforward control is triggered in advance;

[0047] Based on the personalized utility function, a candidate output result matching the current user state is selected from multiple candidate control parameter adjustment instructions;

[0048] The candidate output results are used as the control parameter adjustment instructions.

[0049] Preferably, in step S6, the step of controlling the air conditioning equipment to execute the target control parameter adjustment command includes:

[0050] Establish an optimization objective based on constraints of thermal comfort index and air conditioning equipment operating costs;

[0051] A Gaussian process is used as a surrogate model to estimate the target results corresponding to different control parameter adjustment commands;

[0052] The control parameter adjustment command to be evaluated is selected based on the acquisition function;

[0053] When the preset convergence condition is met, the current optimal control parameter adjustment command is output.

[0054] Preferably, after step S6, the method further includes:

[0055] Continuously acquire millimeter-wave radar data and infrared thermal imaging data after the air conditioning equipment executes the target control parameter adjustment command;

[0056] Update the corresponding physiological and environmental parameter data;

[0057] Based on the updated physiological parameter data, environmental parameter data, and user feedback or user behavior information, the individual physiological benchmark library and the personalized utility function are updated;

[0058] Based on the updated results, steps S3 to S6 are executed again to form a closed-loop personalized control process.

[0059] The present invention has the following beneficial effects:

[0060] 1. This invention acquires a user's heart rate, respiratory rate, and body surface temperature distribution in a non-contact manner using millimeter-wave radar and infrared thermal imaging. It uses the above physiological parameters and environmental parameters as input to predict the trend of thermal comfort index changes. When it is predicted that the thermal comfort index is at risk of falling to a preset threshold, it generates control parameter adjustment instructions in advance. This transforms indoor environmental control from a method that relies solely on feedback from environmental sensors to an active control method that integrates human physiological state, solving the problem that the thermal comfort state of the human body is not accurately reflected in the prior art due to the adjustment based solely on environmental parameters.

[0061] 2. This invention establishes an individual physiological benchmark library and constructs a personalized utility function, enabling the control strategy to be adjusted according to the user's physiological characteristics and historical behavior, thereby solving the problem that the uniform control method in the prior art cannot reflect individual differences.

[0062] 3. This invention uses an LSTM-Attention temporal prediction network to predict the future trend of thermal comfort index changes. When there is a risk of a decline in the thermal comfort index, it triggers feedforward control in advance, avoiding the lag of traditional feedback control. At the same time, it combines Bayesian optimization to optimize the control parameter adjustment instructions, taking into account the operating energy consumption of air conditioning equipment while meeting the thermal comfort index constraints. This solves the problems of lag in adjustment and difficulty in balancing comfort and energy consumption in the prior art, and realizes predictive and energy-saving intelligent control of indoor environment. Attached Figure Description

[0063] Figure 1 This is a flowchart of the method for personalized indoor environment control that integrates human physiological parameters, as proposed in this invention. Detailed Implementation

[0064] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] Reference Figure 1 This invention provides a method for personalized indoor environment control that integrates human physiological parameters, comprising the following steps:

[0066] S1. Acquire millimeter-wave radar data and infrared thermal imaging data of indoor users, and extract heart rate and respiratory rate based on millimeter-wave radar data, and extract body surface temperature distribution based on infrared thermal imaging data.

[0067] S2. Establish an individual physiological baseline database corresponding to the user based on heart rate, respiratory rate and body surface temperature distribution, and construct a personalized utility function based on the individual physiological baseline database. The personalized utility function is a decision model used to characterize the user's preference for different environmental conditions and different air conditioning equipment operating parameters.

[0068] S3. Input the physiological parameter data, environmental parameter data and air conditioning equipment operation parameters of the current time and historical time into the time series prediction network to predict the trend of physiological parameter change and thermal comfort index change in the future. The thermal comfort index is a comprehensive thermal comfort evaluation index calculated based on physiological parameter data and environmental parameter data.

[0069] S4. The prediction process of the time series prediction network is constrained by the physical information neural network, which uses the Fanger thermal comfort equation as the physical constraint.

[0070] S5. When the predicted thermal comfort index is at risk of falling to a preset threshold, control parameter adjustment instructions for the air conditioning equipment are generated based on the personalized utility function to achieve feedforward control.

[0071] S6. Optimize the control parameter adjustment command through Bayesian optimization, determine the target control parameter adjustment command under the premise of meeting comfort constraints, and control the air conditioning equipment to execute the target control parameter adjustment command.

[0072] Specifically, in step S1, a millimeter-wave radar device and an infrared thermal imaging device are deployed in the indoor environment to perform non-contact sensing of users in the indoor space. The millimeter-wave radar continuously emits electromagnetic wave signals and receives the reflected echoes from the human body. By filtering the echo signals by distance dimension, the main reflection area corresponding to the human body is determined. The echo signals from the main reflection area are processed by phase unfolding and trend term elimination to obtain a phase sequence characterizing the periodic micro-movement changes of the thoracic cavity. The phase sequence is then processed by frequency band separation to obtain the respiratory signal component and the cardiac signal component. Spectral analysis is performed on the respiratory signal component and the cardiac signal component to determine the corresponding respiratory rate and heart rate. Simultaneously, the infrared thermal imaging device acquires thermal radiation images of the human body surface, identifies the human body region in the thermal radiation image, and obtains the user's body surface region. The average body surface temperature within the body surface region is statistically analyzed, and the degree of body surface temperature change is calculated to obtain body surface temperature distribution information. The heart rate, respiratory rate, and body surface temperature distribution are time-aligned to form physiological parameter data corresponding to a unified time.

[0073] In step S2, physiological parameter data collected from the same user over multiple time periods are statistically processed to establish an individual physiological benchmark database, which characterizes the range of physiological characteristics of the user under stable conditions. Based on the degree of difference between the current physiological parameter data and the individual physiological benchmark database, the changes in the user's physiological state are determined. On this basis, a personalized utility function model is constructed to describe the influence of different indoor environmental conditions and different air conditioning equipment operating parameters on the degree of matching of the user's physiological state. By recording the user's behavioral changes or feedback information under different control strategies, the personalized utility function is continuously updated to reflect the changes in the user's preferences during long-term use.

[0074] In step S3, physiological parameter data, indoor environmental parameter data, and air conditioning equipment operating parameters from multiple consecutive time points are constructed into a time series input sample according to time sequence and input into the time series prediction network. The time series prediction network models the time dependency relationship between historical data and outputs the physiological parameter change trend and thermal comfort index change trend for several future time points. The thermal comfort index is a comprehensive evaluation index reflecting the current thermal state of the user under specific environmental conditions, which is determined by both physiological parameter data and environmental parameter data.

[0075] In step S4, a physical constraint model for thermal comfort is established based on the human body's thermal balance mechanism, and this physical constraint model is embedded into the training process of the time-series prediction network. By constraining the deviation between the predicted thermal comfort index and the thermal comfort index calculated based on the thermal comfort mechanism, the prediction model can meet the historical data fitting requirements while conforming to the thermal comfort mechanism relationship, thereby obtaining the thermal comfort trend prediction result constrained by physical information.

[0076] In step S5, the future thermal comfort index change trend output by the time-series prediction network is judged. When the prediction result indicates that there is a risk that the thermal comfort index will drop to outside the preset comfort range boundary at a future time, feedforward control is triggered in advance while the current thermal comfort index is still within the comfort range. Based on the personalized utility function, a combination of control parameters that matches the current user's physiological state is selected from the preset candidate control strategy set, and the corresponding air conditioning equipment control parameter adjustment command is generated.

[0077] In step S6, an optimization problem is established for the control parameter adjustment command with thermal comfort index constraints and equipment operation consumption constraints as objectives. Bayesian optimization method is used to iteratively evaluate and update different combinations of control parameters. The target result is estimated through a surrogate model, and a new combination of control parameters is selected for search based on the acquisition strategy. When the preset convergence condition is met, the current optimal target control parameter adjustment command is output and sent to the air conditioning equipment control terminal for execution, thereby completing an indoor environment regulation process based on the changing trend of human physiological state.

[0078] Furthermore, in step S1, the steps of extracting heart rate and respiratory rate based on millimeter-wave radar data include:

[0079] Target range gates are used to select millimeter-wave radar data in order to determine the main reflection area corresponding to indoor users;

[0080] Phase unwrapping and detrending processing are performed on the echo signal corresponding to the main reflection area to obtain a phase sequence characterizing the micro-motion changes of the thoracic cavity;

[0081] Doppler filtering and frequency band separation were performed on the phase sequence to obtain the respiratory signal component and the heartbeat signal component, respectively.

[0082] Spectral peak detection was performed on the respiratory and cardiac signal components to determine the respiratory rate and heart rate, respectively.

[0083] In step S1, the step of extracting the body surface temperature distribution based on infrared thermal imaging data includes:

[0084] Human body region segmentation is performed on infrared thermal imaging data to obtain the corresponding body surface region of indoor users;

[0085] Extract the average body surface temperature and body surface temperature difference features based on the body surface region;

[0086] The average body surface temperature and the characteristics of body surface temperature difference are used as the characterization results of body surface temperature distribution.

[0087] The distribution of body surface temperature is time-aligned with heart rate and respiratory rate to form physiological parameter data corresponding to the same moment.

[0088] Specifically, during the millimeter-wave radar acquisition process, the radar equipment continuously scans the indoor space according to a preset sampling period, obtaining a time-varying echo signal sequence. A discrete Fourier transform is performed on the echo signals in the range dimension to obtain the complex echo spectrum corresponding to different range cells. By calculating the cumulative energy value of each range cell within the time window, the range cell with the highest energy is determined as the primary reflection area of ​​the user, thus completing the target range gate selection.

[0089] The original phase sequence is obtained by performing an arctangent operation on the complex echo signal corresponding to the main reflection region. Let the complex echo signal be... ;in, Indicates in-phase components, Indicates orthogonal components, If the imaginary unit is represented, then the corresponding phase sequence satisfies:

[0090] ;

[0091] in, Indicates time The original phase value is obtained. Due to the periodic jump in phase, the original phase sequence is unwrapped and low-frequency trend terms are removed by polynomial fitting to obtain a continuously changing phase sequence, which is used to characterize the micro-motion changes of the thoracic cavity under the action of respiration and heartbeat.

[0092] A frequency band separation process is performed on the continuous phase sequence. By constructing bandpass filters with different passband ranges, the low-frequency component is treated as the respiratory signal component, and the high-frequency component as the heartbeat signal component. A Fast Fourier Transform is then performed on the separated signal to obtain the corresponding spectral distribution. Let the spectral amplitude function be... Then, the frequency value corresponding to the maximum amplitude within the preset frequency range is found and used as the respiratory rate and heart rate, respectively. Further, the respiratory rate and heart rate are obtained based on the correspondence between frequency and the number of cycles per unit time.

[0093] In infrared thermal imaging data processing, human body regions are identified through image thresholding and morphological operations, resulting in a pixel set of the body surface region. Statistical processing of the temperature values ​​within this pixel set yields the average body surface temperature and the degree of dispersion of body surface temperature. Let the pixel set of the body surface region be... Then the average body surface temperature satisfies:

[0094] ;

[0095] in, Indicates time The average body surface temperature Represents pixels Temperature value, This represents the number of pixels in the body surface region. To characterize the degree of change in body surface temperature distribution, the temperature dispersion of the body surface region is calculated to obtain the body surface temperature difference characteristics:

[0096] ;

[0097] in, It represents a characteristic quantity of body surface temperature distribution.

[0098] After extracting the heart rate, respiratory rate, and body surface temperature distributions, time synchronization processing is performed on the data from different sensors. By unifying the sampling timestamps or using interpolation, data from different sampling frequencies are mapped to the same time series, thereby forming a complete physiological parameter data vector.

[0099] Through the above processing flow, continuous physiological state change information can be obtained without contacting the user, and a unified physiological parameter dataset can be formed for subsequent modeling and prediction, providing basic input for personalized indoor environment control.

[0100] Furthermore, in step S2, the step of establishing an individual physiological baseline library corresponding to the user based on heart rate, respiratory rate, and body surface temperature distribution includes:

[0101] Collect user heart rate, respiratory rate and body surface temperature distribution at multiple time periods to form an individual baseline sample set;

[0102] An individual physiological benchmark library is established based on an individual benchmark sample set to characterize the physiological characteristics of users in a resting state and their fluctuation range.

[0103] Personalized utility functions are constructed based on individual physiological benchmark databases and the historical physiological parameter change trends of users.

[0104] Personalized utility functions are used to evaluate preferences for different environmental conditions and different combinations of air conditioning equipment operating parameters.

[0105] In step S2, the step of constructing the personalized utility function includes:

[0106] Establish a set of candidate control strategies, with different candidate control strategies corresponding to different combinations of air conditioning equipment operating parameters;

[0107] Reward information is determined based on user feedback or behavioral changes after the implementation of candidate control strategies;

[0108] The algorithm for multi-armed slot machines is used to learn reward information in order to determine the preference strategies that match the user.

[0109] The personalized utility function is updated based on the preference strategy.

[0110] Specifically, during the establishment of the individual physiological baseline database, physiological parameters collected from users at different time periods are uniformly organized. Let the first... The physiological parameter vector obtained at each acquisition time is:

[0111] ;

[0112] in, Indicates heart rate, Indicates respiratory rate, This represents the average body surface temperature. It represents a characteristic quantity of body surface temperature distribution.

[0113] Statistical processing of the individual baseline sample set yields the individual baseline mean vector: ;in, This represents the vector of individual physiological baseline means. This represents the sample size. Simultaneously, the covariance matrix is ​​calculated for the sample set:

[0114] ;

[0115] in, A statistical description matrix representing the range of fluctuations in an individual's physiological parameters.

[0116] The degree of deviation of the current physiological state is expressed by the following formula: ;in, This represents the vector of physiological parameters at the current moment. It represents the offset of the current physiological state relative to the individual's baseline state.

[0117] A personalized utility function is constructed based on the offset and historical physiological change trends. This personalized utility function measures the degree to which different control strategies match the current user's physiological state, and its form can be expressed as:

[0118] ;

[0119] in, Indicates the first A candidate control strategy at time 1 The utility function value, This indicates the current thermal comfort index. Indicates the target thermal comfort index. The L2 norm of the offset vector, Indicates the first The control cost corresponding to each control strategy This represents the weighting coefficient.

[0120] During the establishment of the candidate control strategy set, the supply air temperature, supply air velocity, and supply air volume of the air conditioning equipment are combined according to a preset discrete step length to form multiple control parameter schemes. Each control parameter scheme corresponds to a candidate control strategy. When a candidate control strategy is executed, a reward value is generated by recording the changes in the thermal comfort index and user behavior information after execution.

[0121] The user's feedback or behavioral changes after executing a candidate control strategy can be determined in various ways to generate reward information. For example, in explicit feedback, the user manually intervenes in the environment via a smart terminal app or remote control, such as raising or lowering the set temperature or adjusting the fan speed. If the user does not perform any manual intervention within a preset observation time window (e.g., 30 minutes) after the candidate control strategy is executed, a positive reward (e.g., +1) is assigned; if the user performs manual intervention within the observation time window, a negative or smaller reward value is assigned based on the direction and magnitude of the intervention (e.g., intervention opposite to the current control strategy is assigned -1). In implicit feedback, the user's indoor state changes are monitored using millimeter-wave radar or infrared thermal imaging. If the user remains indoors after the candidate control strategy is executed, the current environmental control effect is considered to meet their comfort needs, and a reward of +0.5 is assigned; if the user leaves the room within the preset observation time window, and the predicted thermal comfort index deviates from the target range before leaving, the current control strategy is considered to have failed to maintain their comfort state, and a reward of -0.5 is assigned. The explicit and implicit feedback mentioned above can be used in combination to form a reward signal.

[0122] Let the first The control strategy in the first The reward obtained after this execution is Then the average reward value of this control strategy satisfies:

[0123] ;

[0124] in, This represents the average reward estimate for the candidate control policy.

[0125] Based on the multi-armed slot machine algorithm, candidate control strategies that meet the following conditions are selected in each control decision process:

[0126] ;

[0127] in, Indicates the index of the currently selected control strategy. Indicates the exploration coefficient. Indicates control strategy The number of executions is determined. Based on the selection results, the weight parameters in the personalized utility function are updated, gradually aligning the control strategy selection process with the user's physiological characteristics.

[0128] Specifically, the weighting coefficients The update process is as follows: Each time a control policy is executed and a reward is obtained... Then, the reward value is compared with the utility value predicted by the current utility function to calculate the prediction error. Using gradient descent or a similar optimization algorithm, the weight coefficients are fine-tuned to minimize the prediction error. After multiple iterations, the weight coefficients gradually converge to values ​​that accurately reflect the user's true preferences. Through this online learning mechanism, the personalized utility function can adaptively approximate the user's comfort preferences without manual calibration.

[0129] Through the above-mentioned individual physiological modeling and preference learning process, an individual physiological benchmark library reflecting the characteristics of changes in the user's physiological state can be formed, and the personalized utility function can be gradually adjusted during continuous control operation, so that subsequent environmental regulation decisions are executed based on the user's own physiological characteristics.

[0130] Furthermore, in step S3, the steps of predicting the future trends of physiological parameter changes and thermal comfort index changes include:

[0131] Construct a time series input sample consisting of physiological parameter data, environmental parameter data, and air conditioning equipment operating parameters from multiple consecutive time points;

[0132] Input the time-series input samples into the LSTM-Attention network;

[0133] Extracting temporal dependency features from time series using an LSTM network;

[0134] The features at different historical moments are weighted using the Attention mechanism;

[0135] Based on the weighted temporal features, the output shows the future trends of physiological parameter changes and thermal comfort index changes.

[0136] Specifically, during the construction of the time series input samples, the physiological parameter vectors, environmental parameter vectors, and air conditioning operation parameters from the previous control cycle at each time point are concatenated to form an input feature vector of uniform dimension. Let time point be... The input feature vector is:

[0137] ;

[0138] in, This represents a vector of physiological parameters, including heart rate, respiratory rate, and body surface temperature distribution characteristics. This represents a vector of environmental parameters, including air temperature, relative humidity, air velocity, and mean radiant temperature. This represents the vector of air conditioning operating parameters from the previous control cycle.

[0139] Within a time window length of Under these conditions, the input feature vectors from multiple consecutive time points are combined to form a time series input sample: ;in, This represents the set of time series samples used for prediction.

[0140] In the feature extraction process of the LSTM network, time-series input samples are recursively calculated step by step. Let... Indicates time The hidden state vector, Let the memory cell state vector be represented, then the LSTM state update relationship can be expressed as: ;in, Represents the output gate vector, symbol This represents element-wise vector multiplication. (Memory unit state) By updating historical states through forgetting and input gates, the temporal dependencies between physiological and environmental changes are modeled. After recursive calculation of all historical moments, the hidden state sequence is obtained. The hidden state sequence is used for subsequent Attention weight calculation.

[0141] In one specific implementation of this embodiment, the LSTM-Attention network can adopt the following structure: the LSTM layer consists of two layers, each with 64 hidden units; the Attention layer uses a single-head self-attention mechanism, and the dimensions of the query vector, key vector, and value vector are all set to 32; the output layer is a fully connected layer, and the output dimension is equal to the prediction stride. Multiply by the number of variables to be predicted. The above network structure is merely an illustrative example. Those skilled in the art should understand that, depending on the computational resources and accuracy requirements of specific application scenarios, hyperparameters such as the number of network layers, the number of hidden units, and the number of attention heads can be adaptively adjusted, and these adjustments do not depart from the protection scope of this invention.

[0142] During the Attention weight calculation process, a relevance score is calculated for the hidden state at each historical time step. Let the i-th... The rating of each historical moment is Then the normalized weights satisfy:

[0143] ;

[0144] in, Indicates the first The contribution weight of each historical moment feature to the current prediction task.

[0145] The hidden state sequence is weighted and summed based on the weights to obtain the comprehensive temporal feature vector. ;in, This represents the temporal feature vector after Attention weighting.

[0146] The comprehensive temporal feature vector is input into the prediction output layer to obtain the trends of physiological parameter changes and thermal comfort index changes at multiple future time points. Let the prediction step size be... Then the prediction result satisfies:

[0147] ;

[0148] in, Indicates the future number The predicted output vector at each time point contains predicted physiological parameter values ​​and predicted thermal comfort index values. This indicates the output mapping function.

[0149] Through the above-mentioned time-series prediction modeling process, the temporal correlation between changes in human physiological state and environmental changes is modeled, enabling subsequent control decisions to be executed based on future trend information, thereby forming regulatory input data based on physiological state prediction results.

[0150] Furthermore, in step S4, the step of constraining the prediction process of the time series prediction network using a physical information neural network includes:

[0151] Obtain the thermal comfort index prediction results output by the time-series prediction network;

[0152] Based on physiological and environmental parameter data, the physical constraint results of the thermal comfort index are calculated using the Fanger thermal comfort equation.

[0153] The model training process is corrected based on the deviation between the predicted thermal comfort index and the physical constraint results of the thermal comfort index.

[0154] The training of the physical information neural network is completed by jointly optimizing the data fitting term and the physical constraint term.

[0155] Specifically, in the output stage of the prediction model, the predicted value of the thermal comfort index obtained from the time-series prediction network is used as the data-driven prediction result. Let time... The prediction results are as follows: ;in, This represents the predicted value of the thermal comfort index. This represents the time series input sample. Indicates by parameters The representation of the prediction network mapping relationship.

[0156] In the physical constraint calculation process, based on the human body's thermal balance mechanism, the thermal comfort index under physical constraints is calculated using the Fanger thermal comfort equation. Let the physical constraint result for the thermal comfort index be:

[0157] ;

[0158] in, This represents the index value calculated from the thermal comfort mechanism. Indicates the human metabolic rate. Indicates air temperature. This represents the average radiant temperature. Indicates air velocity. Indicates the partial pressure of water vapor. Indicates thermal resistance when wearing clothing. This represents the mapping function of thermal comfort mechanism.

[0159] In practical implementation, the human metabolic rate The data is calculated from the physiological parameters. Based on the well-known estimation method for the relationship between heart rate and metabolic rate in the ISO 8996 standard, a mapping relationship from heart rate to metabolic rate can be established. This embodiment preferably uses the following mapping method:

[0160] ;

[0161] in, The basal metabolic rate of the user at rest (e.g., a possible value for an office setting). ), Current heart rate, The resting heart rate (which can be derived from the mean vector of an individual's physiological baseline database) (obtained from) The user's maximum heart rate (which can be estimated as "220 - age"). To and The corresponding maximum metabolic rate. In cases where personal information such as user age is unavailable, a simplified linear regression model can be used as an alternative: , where the coefficient and The mapping can be determined through pre-experimental calibration or based on population statistics. Regardless of the mapping method used, heart rate data can be converted into metabolic rate inputs that can be used for calculations of the Fanger thermal comfort equation.

[0162] thermal resistance of clothing This can be determined in one of the following ways: 1) Based on the current season and indoor temperature, the system presets a typical dress code (for example, typical summer office dress code). Take in winter ); 2) If the system has user clothing recognition capabilities (e.g., through infrared thermal imaging or visible light image analysis), the recognition results can be mapped to the corresponding thermal resistance value; 3) As a simplified solution, the system allows users to input or select their clothing status through an interactive interface to obtain more accurate information. value.

[0163] The partial pressure of water vapor is obtained by converting relative humidity and air temperature. The average radiant temperature is determined by indoor thermal radiation measurements.

[0164] To impose physical constraints on the prediction model, a prediction bias is introduced:

[0165] ;

[0166] in, This represents the difference between the data-driven prediction result and the physical constraint result. During model training, a joint loss function is constructed simultaneously, consisting of a data fitting term and a physical constraint term. ;in, Represents the overall loss function. This represents the error term between the predicted value and the actual observed value. This represents the error term between the predicted value and the physical constraint result. This represents the weighting coefficients. The network parameters are iteratively updated by minimizing the overall loss function, ensuring that the prediction results satisfy both the consistency of the observed data and the constraints of the thermal comfort mechanism.

[0167] In this embodiment, the physical constraints of the physical information neural network are applied to the training phase of the temporal prediction network. That is, during the training process of the LSTM-Attention network, the physical constraint terms in the joint loss function are applied. Constraints are imposed on network parameter updates to ensure that the output of the trained time-series prediction network simultaneously meets the requirements of data fitting and thermal comfort mechanism. Specifically, in each training iteration, the predicted value of the thermal comfort index output by the time-series prediction network is first calculated based on the current network parameters. Simultaneously, the physical constraint results were calculated using the Fanger thermal comfort equation. Then calculate the joint loss function. The parameters of the LSTM-Attention network are simultaneously updated using the backpropagation algorithm until the loss function converges. This joint training method internalizes the physical constraints into the network parameters, allowing them to be directly used for inference and prediction after training without additional computational steps.

[0168] During model operation, the predicted thermal comfort index, corrected for physical constraints, is used as input for subsequent control decisions. Through the aforementioned physical information neural network training method, the output of the prediction model is made numerically consistent with the human body's thermal balance mechanism, thus forming a predictive data foundation that meets the needs of indoor environmental regulation.

[0169] Furthermore, in step S5, the step of generating control parameter adjustment instructions for the air conditioning equipment based on the personalized utility function includes:

[0170] Based on the future trend of thermal comfort index changes, determine whether there is a risk of it dropping to a preset threshold;

[0171] Trigger feedforward control in advance when risks exist;

[0172] Based on the personalized utility function, a candidate output result matching the current user state is selected from multiple candidate control parameter adjustment instructions.

[0173] The candidate output results are used as control parameter adjustment instructions.

[0174] Specifically, after the prediction model outputs a series of thermal comfort indices for multiple future time periods, the prediction results are processed to determine intervals. Let the prediction step size be... The predicted thermal comfort index sequence is represented as: ;in, Indicates the future number The predicted thermal comfort index at each time point. Let the boundary of the comfort interval be... In this embodiment, the comfort zone boundaries can be set with reference to the PMV comfort range recommended in international standards ASHRAE Standard 55 or ISO 7730. For example, for a typical office environment, the boundaries can be set as follows: Set as , Set as This corresponds to a thermal environment that satisfies over 90% of people. Of course, this threshold range can be customized by the system administrator or the user based on specific application scenarios (such as sleep environments, sports venues, etc.) and individual user needs. This invention does not impose strict limitations on this. When: or At that time, it is determined that there is a risk that the future thermal comfort index will deviate from the preset threshold.

[0175] After completing the risk assessment, the control strategy generation stage begins. First, a set of candidate control parameter adjustment instructions is established. Let the... The air conditioning operating parameter vector corresponding to the candidate control parameter adjustment commands is:

[0176] ;

[0177] in, This indicates the supply air temperature setpoint. This indicates the air supply speed setting value. This indicates the set value for the air supply volume.

[0178] At the current moment, the utility of each candidate control parameter adjustment command is evaluated. Let the personalized utility function affect the first... The evaluation results of the candidate control parameter adjustment commands are as follows: ;in, Indicates the first The utility function value corresponding to each control command. This indicates the predicted thermal comfort index at the next moment after the control command is executed. Indicates the target thermal comfort index. This represents the offset vector of the current physiological state. This indicates the control cost corresponding to executing the control instruction. This represents the weighting coefficient.

[0179] Based on the utility function evaluation results, select the control parameter adjustment instructions that meet the requirements from the candidate control parameter adjustment instruction set. The control command is used as the current control output result, where Indicates the selected control instruction index.

[0180] After selecting the control command, the control parameter adjustment command is sent to the air conditioning control execution unit to adjust the supply air temperature, supply air speed, and supply air volume. By generating and executing the control command before the thermal comfort index actually deviates from the comfort range, the environmental regulation process responds in advance based on the prediction results, thus forming a control decision-making process based on the trend of physiological state changes.

[0181] Furthermore, in step S6, the step of controlling the air conditioning equipment to execute the target control parameter adjustment command includes:

[0182] Establish an optimization objective based on constraints of thermal comfort index and air conditioning equipment operating costs;

[0183] A Gaussian process is used as a surrogate model to estimate the target results corresponding to different control parameter adjustment commands;

[0184] The control parameter adjustment command to be evaluated is selected based on the acquisition function;

[0185] When the preset convergence condition is met, the current optimal control parameter adjustment command is output.

[0186] Specifically, in the process of constructing the optimization objective, the predicted thermal comfort index and the operating cost of the air conditioning equipment are modeled as joint optimization factors. Let the control parameter vector be: ;in, This indicates the supply air temperature setpoint. This indicates the air supply speed setting value. This indicates the set value for the air supply volume.

[0187] With a prediction step size of Under the given conditions, construct the optimization objective function:

[0188] ;

[0189] in, This represents the comprehensive evaluation value corresponding to the control parameters. Indicates the future number Predicted thermal comfort index values ​​at each moment. Indicates the target thermal comfort index. Represents the control cost function. This represents the cost weighting coefficient. The control cost function is calculated based on the operating power and operating time of the air conditioning equipment and is used to reflect the equipment operating costs under different combinations of control parameters.

[0190] In Bayesian optimization, the objective function is treated as an unknown mapping relationship, and a surrogate model is established using a Gaussian process. Let the control parameter vector be the input variable; then the probability distribution of the objective function satisfies... ;in, Represents the mean function, Represents the covariance function, symbol This represents the distribution of a Gaussian process.

[0191] After obtaining several historical evaluation samples, the Gaussian process can provide the posterior mean of the adjustment command for any candidate control parameter. With posterior standard deviation These are used to characterize the estimation result and uncertainty of the objective function value corresponding to the control parameter combination, respectively. During the acquisition function calculation phase, based on the current optimal objective function value... Construct the desired improvement metrics. Let:

[0192] ;

[0193] The corresponding acquisition function can then be expressed as:

[0194] ;

[0195] in, Indicates the expected improvement value. Represents the standard normal distribution function. This represents the standard normal probability density function. By performing an extremum search on the acquisition function, the next set of control parameter adjustment commands to be evaluated is determined. After completing the objective function evaluation, the new samples are added to the historical sample set to update the Gaussian process model.

[0196] When the change in the optimal value of the objective function during continuous iteration satisfies When the optimization process reaches the convergence condition, the following conditions are met: Indicates the first The optimal objective function value in the next iteration. This indicates a preset convergence threshold. Once the convergence condition is met, the corresponding control parameter adjustment command is sent as the target control parameter adjustment command to the air conditioning equipment for execution. The air conditioning equipment adjusts the supply air temperature, supply air velocity, and supply air volume according to the target control parameter adjustment command, thereby completing the environmental control process.

[0197] Through the above-mentioned optimized decision-making and execution process, the control parameter selection process is jointly determined based on the predicted thermal comfort index and equipment operating constraints, thereby forming an environmental regulation control output that can be used for continuous operation.

[0198] Furthermore, after step S6, the following steps are also included:

[0199] Continuously acquire millimeter-wave radar data and infrared thermal imaging data after the air conditioning equipment executes the target control parameter adjustment command;

[0200] Update the corresponding physiological and environmental parameter data;

[0201] Based on updated physiological parameter data, environmental parameter data, and user feedback or user behavior information, the individual physiological benchmark database and personalized utility function are updated.

[0202] Based on the updated results, steps S3 to S6 are executed again to form a closed-loop personalized control process.

[0203] Specifically, after the air conditioning equipment executes the target control parameter adjustment command, it continuously monitors the physiological state of indoor users according to a preset sampling period, obtaining new millimeter-wave radar echo signals and infrared thermal imaging data. Then, following the aforementioned steps, it extracts updated heart rate, respiratory rate, and body surface temperature distribution characteristics. Simultaneously, it acquires air temperature, relative humidity, air velocity, and mean radiant temperature after the control command is executed through environmental sensors, thereby forming an updated environmental parameter vector.

[0204] Let the updated physiological parameter vector be:

[0205] ;

[0206] in, This indicates the updated heart rate. This indicates the updated respiratory rate. This represents the updated average body surface temperature. This represents the updated body surface temperature distribution characteristic.

[0207] During the update of the individual physiological baseline database, newly collected physiological parameter vectors are added to the historical sample set, and the individual baseline mean is updated recursively. Let the original baseline mean be... The sample size is Then the updated benchmark mean satisfies:

[0208] ;

[0209] in, This represents the updated vector of individual physiological baseline means.

[0210] During the personalized utility function update process, a reward value is generated based on the change in the thermal comfort index after executing the control command and user behavior feedback. Let the reward corresponding to this control execution be... Then the cumulative reward estimate of the control strategy satisfies:

[0211] ;

[0212] in, Indicates control strategy At any moment The average reward estimate, This indicates the number of times the control strategy has been executed. Based on the updated reward estimate, the weight parameters in the personalized utility function are adjusted to ensure the utility function reflects the correspondence between the user's current physiological state and environmental changes. After updating the baseline library and utility function, the updated physiological and environmental parameter data are re-input into the time-series prediction network to predict future trends in the thermal comfort index. New control parameter adjustment instructions are then determined according to the aforementioned control strategy generation and optimization process.

[0213] Through the aforementioned continuous updating and cyclical execution mechanism, the indoor environmental control process forms a closed-loop operation mode based on real-time physiological state changes, thereby achieving a continuous control process for the same user's state changes.

[0214] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for personalized indoor environment control that integrates human physiological parameters, characterized in that, Includes the following steps: S1. Acquire millimeter-wave radar data and infrared thermal imaging data of indoor users, and extract heart rate and respiratory rate based on the millimeter-wave radar data, and extract body surface temperature distribution based on the infrared thermal imaging data. S2. Based on the heart rate, respiratory rate and body surface temperature distribution, establish an individual physiological benchmark library corresponding to the user, and construct a personalized utility function based on the individual physiological benchmark library. The personalized utility function is a decision model used to characterize the user's preference for different environmental states and different air conditioning equipment operating parameters. S3. Input the physiological parameter data, environmental parameter data and air conditioning equipment operation parameters of the current time and historical time into the time series prediction network to predict the trend of physiological parameter change and thermal comfort index change in the future time. The thermal comfort index is a comprehensive thermal comfort evaluation index calculated based on the physiological parameter data and the environmental parameter data. S4. The prediction process of the time series prediction network is constrained by a physical information neural network, wherein the physical information neural network uses the Fanger thermal comfort equation as a physical constraint. S5. When the predicted thermal comfort index is at risk of falling to a preset threshold, control parameter adjustment instructions for the air conditioning equipment are generated according to the personalized utility function to achieve feedforward control. S6. Optimize the control parameter adjustment command through Bayesian optimization, determine the target control parameter adjustment command under the premise of satisfying comfort constraints, and control the air conditioning equipment to execute the target control parameter adjustment command.

2. The method for personalized indoor environment control incorporating human physiological parameters according to claim 1, characterized in that, In step S1, the step of extracting heart rate and respiratory rate based on the millimeter-wave radar data includes: The millimeter-wave radar data is subjected to target range gating to determine the main reflection area corresponding to the indoor user; Phase unwrapping and detrending processing are performed on the echo signal corresponding to the main reflection region to obtain a phase sequence characterizing the micro-motion changes of the thoracic cavity; The phase sequence was subjected to Doppler filtering and frequency band separation to obtain the respiratory signal component and the heartbeat signal component, respectively. The respiratory signal component and the heartbeat signal component are subjected to spectral peak detection to determine the respiratory rate and the heart rate, respectively.

3. The method for personalized indoor environment control incorporating human physiological parameters according to claim 1, characterized in that, In step S1, the step of extracting the body surface temperature distribution based on the infrared thermal imaging data includes: The infrared thermal imaging data is segmented into human body regions to obtain the body surface regions corresponding to indoor users; Based on the aforementioned body surface region, the average body surface temperature and body surface temperature difference features are extracted. The average body surface temperature and the body surface temperature difference characteristics are used as the characterization results of the body surface temperature distribution; The body surface temperature distribution is time-aligned with the heart rate and respiratory rate to form physiological parameter data corresponding to the same moment.

4. The method for personalized indoor environment control incorporating human physiological parameters according to claim 1, characterized in that, In step S2, the step of establishing an individual physiological baseline database corresponding to the user based on the heart rate, the respiratory rate, and the body surface temperature distribution includes: The heart rate, respiratory rate and body surface temperature distribution of the user were collected at multiple time periods to form an individual baseline sample set; The individual physiological benchmark library is established based on the individual benchmark sample set to characterize the user's physiological characteristics and fluctuation range in a resting state; The personalized utility function is constructed based on the individual physiological benchmark library and the user's historical physiological parameter change trends. The personalized utility function is used to evaluate preferences for different environmental conditions and different combinations of air conditioning equipment operating parameters.

5. The method for personalized indoor environment control integrating human physiological parameters according to claim 1, characterized in that, In step S2, the step of constructing the personalized utility function includes: Establish a set of candidate control strategies, with different candidate control strategies corresponding to different combinations of air conditioning equipment operating parameters; Reward information is determined based on the user's feedback or behavioral changes after executing the candidate control strategy; The reward information is learned based on the multi-armed slot machine algorithm to determine the preference strategy that matches the user; The personalized utility function is updated according to the preference strategy.

6. The method for personalized indoor environment control incorporating human physiological parameters according to claim 1, characterized in that, In step S3, the steps for predicting the future trends of physiological parameters and thermal comfort index include: Construct a time series input sample consisting of physiological parameter data, environmental parameter data, and air conditioning equipment operating parameters from multiple consecutive time points; The time-series input samples are then fed into the LSTM-Attention network. Extracting temporal dependency features from time series using an LSTM network; The features at different historical moments are weighted using the Attention mechanism; Based on the weighted temporal features, the output shows the future trends of physiological parameter changes and thermal comfort index changes.

7. The method for personalized indoor environment control incorporating human physiological parameters according to claim 1, characterized in that, In step S4, the step of constraining the prediction process of the time series prediction network using a physical information neural network includes: Obtain the thermal comfort index prediction results output by the time-series prediction network; Based on the physiological and environmental parameter data, the physical constraint results of the thermal comfort index are calculated using the Fanger thermal comfort equation. The model training process is corrected based on the deviation between the predicted thermal comfort index and the physical constraint results of the thermal comfort index. The training of the physical information neural network is completed by jointly optimizing the data fitting term and the physical constraint term.

8. The method for personalized indoor environment control integrating human physiological parameters according to claim 1, characterized in that, In step S5, the step of generating control parameter adjustment instructions for the air conditioning equipment based on the personalized utility function includes: Based on the future trend of thermal comfort index changes, determine whether there is a risk of it dropping to a preset threshold; In the event of the aforementioned risk, feedforward control is triggered in advance; Based on the personalized utility function, a candidate output result matching the current user state is selected from multiple candidate control parameter adjustment instructions; The candidate output results are used as the control parameter adjustment instructions.

9. The method for personalized indoor environment control incorporating human physiological parameters according to claim 1, characterized in that, In step S6, the step of controlling the air conditioning equipment to execute the target control parameter adjustment command includes: Establish an optimization objective based on constraints of thermal comfort index and air conditioning equipment operating costs; A Gaussian process is used as a surrogate model to estimate the target results corresponding to different control parameter adjustment commands; The control parameter adjustment command to be evaluated is selected based on the acquisition function; When the preset convergence condition is met, the current optimal control parameter adjustment command is output.

10. The method for personalized indoor environment control integrating human physiological parameters according to claim 1, characterized in that, Following step S6, the following is also included: Continuously acquire millimeter-wave radar data and infrared thermal imaging data after the air conditioning equipment executes the target control parameter adjustment command; Update the corresponding physiological and environmental parameter data; Based on the updated physiological parameter data, environmental parameter data, and user feedback or user behavior information, the individual physiological benchmark library and the personalized utility function are updated; Based on the updated results, steps S3 to S6 are executed again to form a closed-loop personalized control process.

Images