Environmental temperature and humidity adjusting method based on body comfort

Abstract

The invention discloses an environment temperature and humidity adjusting method based on the body comfort degree. The method comprises the following steps that 1, environment parameters and physiological data are collected; 2, dynamic causal diagram model analysis: constructing a dynamic causal diagram model, and identifying a key path; step 3, CFD lightweight simulation: utilizing a lightweight CFD simulation tool to generate a local eddy current model, and dynamically correcting a calculation formula of humidity; 4, sweat evaporation kinetic model calculation: calculating body surface humidity; 5, double-ring reinforcement learning framework adjustment: performing inner-ring double-ring reinforcement learning adjustment: dynamically adjusting temperature and humidity set values by taking the real-time comfort score of the user as an award, and automatically adjusting the environment temperature by double-ring reinforcement learning when the comfort level fed back by the user is reduced, and performing outer-ring double-ring reinforcement learning prediction: predicting the long-term preference of the user in combination with historical data; step 6, performing adversarial security constraint guarantee; and 7, generating and executing an adjusting instruction: generating the adjusting instruction according to the perspiration state of the user and the environmental parameters.

Description

Technical Field

[0001] The present invention relates to the field of smart home technology, and in particular to a method for regulating ambient temperature and humidity based on physical comfort. Background Art

[0002] With the improvement of people's living standards and the continuous advancement of technology, the demand for comfortable living and working environments is also increasing. Ambient temperature and humidity are important factors affecting human comfort, and their regulation is crucial for improving quality of life, work efficiency, and health. Ambient temperature and humidity regulation technology based on human comfort is a key innovation in the smart home field. It drives the development of smart home systems towards greater intelligence and personalization, improving their overall performance and user experience.

[0003] Traditional ambient temperature and humidity control systems and methods usually use fixed control logic or preset parameter settings. These settings may be based on general comfort preferences or industry standards, but often fail to fully account for individual differences and dynamically changing environmental conditions. Users have to manually adjust the settings of the air conditioner or humidifier to seek a more comfortable indoor environment. Therefore, a method for ambient temperature and humidity control based on physical comfort is proposed. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a method for regulating environmental temperature and humidity based on body comfort.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A method for regulating ambient temperature and humidity based on body comfort, comprising the following steps: Step 1: Environmental parameter and physiological data collection: A microenvironmental sensor array is used to collect 3D distribution data of temperature, humidity, and airflow velocity in a specified area in real time. The sensor data is processed by an edge computing layer (such as a Raspberry Pi 4B + Coral TPU) to generate a thermal map of the environmental parameters. A flexible electronic skin patch (with an integrated impedance sensor and a micro sweat pH detection module) is used to obtain skin surface electrophysiological data (such as skin impedance and sweat pH) in real time. The patch data is also processed by the edge computing layer to extract key parameters related to sweat secretion and skin condition, providing input for the sweat evaporation dynamics model. Step 2: Dynamic Causal Graph Model Analysis: Analyze the causal relationship between ambient temperature, ambient humidity, body surface humidity, and body temperature parameters, construct a dynamic causal graph model, identify key paths, and determine the degree of influence of each parameter on the somatosensory index. Use a Bayesian network to update parameter weights in real time, retaining only parameters with a significant impact on the somatosensory index. Step 3: CFD lightweight simulation: Use lightweight CFD simulation tools to generate a local eddy flow model, taking into account the impact of obstacles, human body shape and other factors on airflow. Based on the CFD simulation results, dynamically modify the calculation formula of ambient humidity; Step 4: Calculation of the sweat evaporation dynamics model: Based on parameters such as skin impedance and airflow velocity, a sweat evaporation dynamics model is established. The ambient humidity boundary conditions are updated according to the corrected ambient humidity, and the body surface humidity is calculated. The thermal map of the environmental parameters is spatially and temporally aligned with the physiological data to form a multimodal input vector, which serves as the initial and boundary conditions of the sweat evaporation dynamics model. Step 5: Dual-loop reinforcement learning framework adjustment: The inner-loop dual-loop reinforcement learning (RL) adjustment uses the user's real-time comfort score as a reward to dynamically adjust the temperature and humidity set points. When the user's feedback shows a decrease in comfort, the dual-loop reinforcement learning (RL) automatically adjusts the ambient temperature to quickly respond to user needs. The outer-loop dual-loop reinforcement learning (RL) prediction combines historical data (such as seasons and circadian rhythms) to predict long-term user preferences. Step 6: Adversarial safety constraint assurance: Control the temperature and humidity adjustment within the medical safety range. Introduce a penalty term in double-loop reinforcement learning (RL) to provide a negative reward when the adjustment instruction exceeds the safety range. Step 7: Generate and execute adjustment instructions: Generate adjustment instructions based on user physiological data and environmental parameters, and execute the adjustment instructions through the air conditioning controller and humidifier / dehumidifier.

[0006] The above further includes: Furthermore, the specific steps of constructing the dynamic causal graph model are: Parameter causal relationship analysis: Granger causality test was used to determine the causal relationship between ambient temperature, humidity, surface humidity, body temperature and somatosensory index; Construct a dynamic causal graph model: Based on the causal relationships obtained from the above analysis, use graphical tools to construct a dynamic causal graph model. In the graph, nodes represent each parameter, and directed edges represent the causal relationships between parameters. Identify key paths: Determine the influence weight of each path on the somatosensory index D. Use regression analysis method, take the somatosensory index as the dependent variable, and take ambient temperature, humidity, body surface humidity, and body temperature as independent variables to establish a regression model. The coefficient of the regression model is used to determine the influence degree of each path.

[0007] Furthermore, the specific steps of using the Bayesian network to update parameter weights in real time are: Bayesian network structure learning: Based on the causal relationship between parameters, the Bayesian network structure is preliminarily constructed. The K2 algorithm is used to optimize the preliminarily constructed network structure to find the network structure that can most accurately describe the dependency relationship between parameters. Parameter learning: Based on the determined network structure, the conditional probability distribution of each node in the network is learned by maximum likelihood estimation using the collected data; Parameter weight calculation and screening: Calculate the influence weight of each parameter on the somatosensory index D, measure the correlation between the parameter and the somatosensory index D by calculating the mutual information, set a threshold, and only retain the parameters with significant impact on the somatosensory index D; Model update and optimization: Regularly use new data to re-learn parameters and recalculate weights, update the Bayesian network model, and adapt to environmental changes and dynamic changes in user preferences.

[0008] Furthermore, the specific steps of generating the local eddy current model using the lightweight CFD simulation tool are as follows: Establish an initial model: Determine the spatial scope of the simulation and set the boundary conditions of the space. For example, set the no-slip boundary condition for walls and the airflow exchange conditions at doors and windows based on their opening and closing conditions. Add obstacles (such as furniture) and a geometric model of the human body to the simulation area. The human body can be a simplified 3D model, considering the impact of the main contours and posture of the human body on airflow. Meshing: Use mesh generation tools to divide the simulation area into small grid cells. Local mesh refinement is performed near obstacles and human shapes. For example, near the human body surface, the mesh size can be set to a smaller value, such as 0.05m, while in areas away from the human body, the mesh size can be appropriately increased, such as 0.1m. This ensures that the mesh quality meets the simulation requirements and avoids the occurrence of deformed meshes that affect the accuracy of the simulation results.

[0009] Set physical parameters: Airflow parameters: Set physical parameters such as air density and viscosity, as well as initial conditions such as air velocity and temperature at the inlet; Humidity parameters: Set initial humidity distribution and set the initial humidity value in the room according to actual conditions; Run lightweight CFD simulation: Use lightweight CFD simulation tools, such as a simplified version of OpenFOAM, to perform simulation calculations and solve the Navier-Stokes equations (NS equations) and energy equations to obtain the velocity field, pressure field, and temperature field distribution of the airflow. The NS equations are expressed in rectangular coordinates as ; Among them, u, v, w are the velocity components in the x, y, and z directions respectively, and p is the pressure. is the air density, ν is the kinematic viscosity of air; Analyze simulation results: Extract the distribution information of local eddies from the simulation results, including the location, intensity, and direction of the eddies. The location and intensity of the eddies are determined by analyzing the curl of the velocity field. The curl formula is: ,in, is the curl vector, are the unit vectors in the x, y, and z directions respectively. Observe the changes in airflow around obstacles and human shapes, and analyze their effects on airflow speed, direction, and vortex formation.

[0010] Furthermore, based on the CFD simulation results, the relationship between humidity, air flow velocity and vortex intensity is analyzed to establish the following relationship: ,in, is the corrected humidity value, is the initial humidity value, is the air flow velocity, is the eddy current intensity, and is the correction factor.

[0011] Furthermore, the specific steps of establishing the sweat evaporation dynamics model and calculating the body surface humidity are as follows: Parameter determination: Determine skin impedance, which is acquired in real time by the impedance sensor on the flexible electronic skin patch; determine airflow velocity, which is collected by the microenvironment sensor array; Model establishment: Using the sweat evaporation dynamics model formula ,in, is the body surface humidity, is the temperature difference between core and skin, is physiological data, expressed as skin impedance, is the air flow velocity, and are individual calibration parameters.

[0012] Parameter calibration: Through experiments, we collect the skin impedance, air flow velocity, and body surface humidity data of individuals under different environmental conditions. and Calibration is performed to measure the difference between the model prediction value and the actual measurement value through the model mean square error (MSE), and the model parameters are adjusted through the genetic algorithm to minimize the loss function; Calculation of body surface humidity: Substitute the real-time skin impedance and airflow velocity into the calibrated model formula to calculate the body surface humidity sr.

[0013] Furthermore, the specific steps of the inner-loop dual-loop reinforcement learning (RL) adjustment are: Define the state space: The state space S includes the current ambient temperature, humidity, body surface humidity, body temperature, and somatosensory index; Define the action space: The action space A includes the temperature adjustment amount Δt1 and the humidity adjustment amount ΔE; Define the reward function: The reward function R is based on the user's real-time comfort score S: ; in, Indicates the status of the comfort score, , is the user's current comfort rating, is the user's last comfort rating; Selection strategy: Use Q-learning to select action a based on the current state s to maximize the long-term reward. The Q-learning update formula is expressed as ; in, is the learning rate, is the discount factor, s is the current state, a is the current action, r is the reward, and s' is the next state; Execute actions and update states: After executing action a, observe the new state s' and reward r, and update the Q value or policy parameters.

[0014] Furthermore, the specific steps of the outer-loop dual-loop reinforcement learning (RL) prediction are: Collect historical data: Collect user temperature and humidity preference data in different seasons and circadian rhythms; Define the state space: The state space includes time (season, day and night), historical temperature and humidity setting values, and user feedback; Define the action space: The action space is the long-term temperature and humidity preference prediction value; Define the reward function: The reward function is based on how well the predicted preferences match the actual preferences; Train the prediction model: Use historical data to train the RL model to predict long-term user preferences.

[0015] The present invention has the following beneficial effects: In the present invention, a dual-loop reinforcement learning framework is used to simultaneously achieve long-term prediction of user preferences and real-time comfort optimization. The inner-loop RL dynamically adjusts the temperature and humidity set values according to the user's real-time comfort score, while the outer-loop RL predicts the user's long-term preferences based on historical data, thereby achieving more intelligent adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a step diagram of a method for adjusting ambient temperature and humidity based on body comfort proposed by the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0018] See also Figure 1 As shown, the present invention is a method for regulating environmental temperature and humidity based on body comfort, comprising the following steps: Step 1: Environmental parameter and physiological data collection: A microenvironmental sensor array is used to collect 3D distribution data of temperature, humidity, and airflow velocity in a specified area in real time. The sensor data is processed by an edge computing layer (such as a Raspberry Pi 4B + Coral TPU) to generate a thermal map of the environmental parameters. A flexible electronic skin patch (with an integrated impedance sensor and a micro sweat pH detection module) is used to obtain skin surface electrophysiological data (such as skin impedance and sweat pH) in real time. The patch data is also processed by the edge computing layer to extract key parameters related to sweat secretion and skin condition, providing input for the sweat evaporation dynamics model. Step 2: Dynamic Causal Graph Model Analysis: Analyze the causal relationship between ambient temperature, humidity, surface humidity, and body temperature parameters, construct a dynamic causal graph model, identify key paths, and determine the degree of influence of each parameter on the somatosensory index. Use a Bayesian network to update parameter weights in real time, retaining only parameters with a significant impact on the somatosensory index. Step 3: CFD lightweight simulation: Use lightweight CFD simulation tools to generate a local eddy flow model, taking into account the impact of obstacles, human body shape and other factors on airflow. Based on the CFD simulation results, dynamically modify the humidity calculation formula; Step 4: Calculation of sweat evaporation dynamics model: Based on parameters such as skin impedance and airflow velocity, a sweat evaporation dynamics model is established to calculate body surface humidity. The environmental thermal map is spatiotemporally aligned with the physiological data to form a multimodal input vector, which serves as the initial and boundary conditions of the sweat evaporation dynamics model. Step 5: Dual-loop reinforcement learning framework adjustment: The inner-loop dual-loop reinforcement learning (RL) adjustment uses the user's real-time comfort score as a reward to dynamically adjust the temperature and humidity set points. When the user's feedback shows a decrease in comfort, the dual-loop reinforcement learning (RL) automatically adjusts the ambient temperature to quickly respond to user needs. The outer-loop dual-loop reinforcement learning (RL) prediction combines historical data (such as seasons and circadian rhythms) to predict long-term user preferences. Step 6: Adversarial safety constraint assurance: Control the temperature and humidity adjustment within the medical safety range. Introduce a penalty term in double-loop reinforcement learning (RL) to provide a negative reward when the adjustment instruction exceeds the safety range. Step 7: Generate and execute adjustment instructions: Generate adjustment instructions based on user physiological data and environmental parameters, and execute the adjustment instructions through the air conditioning controller and humidifier / dehumidifier.

[0019] In one embodiment, the specific steps of constructing a dynamic causal graph model are: Parameter causal relationship analysis: Granger causality test was used to determine the causal relationship between ambient temperature, humidity, surface humidity, body temperature and somatosensory index; Construct a dynamic causal graph model: Based on the causal relationships obtained from the above analysis, use graphical tools to construct a dynamic causal graph model. In the graph, nodes represent each parameter, and directed edges represent the causal relationships between parameters. Identify key paths: Determine the influence weight of each path on the somatosensory index D. Use regression analysis method, take the somatosensory index as the dependent variable, and take ambient temperature, humidity, body surface humidity, and body temperature as independent variables to establish a regression model. The coefficient of the regression model is used to determine the influence degree of each path.

[0020] In this embodiment, it is assumed that a set of data is collected in a summer office environment, and the data records are shown in the following table: Among them, the ambient temperature is t1, t2, the humidity is E, the surface humidity is sr, the body temperature is tr, and the somatosensory index is D; Parameter causal relationship analysis: Through observation of the data, it was found that as the ambient temperature t1 increased, the humidity E also gradually increased, showing a positive correlation. It was preliminarily determined that t1 had a causal effect on E. As the humidity E increased, the surface humidity sr also increased, indicating that E had a causal effect on sr. Changes in surface humidity sr and body temperature tr were accompanied by a decrease in the somatosensory index D, indicating that sr and tr jointly affect D. Construct a dynamic causal graph model: Construct the following simple dynamic causal graph model: t1→E→sr, sr→D, tr→D; Identify the critical path: Use a linear regression model to fit the data, , where a, b, c represent the correlation coefficients, and d represents the intercept; The regression equation was obtained through data fitting: D=−0.1×E−0.15×sr−0.05×tr+10.

[0021] It can be seen from the regression equation that the absolute value of the coefficient of sr, 0.15, is relatively large, indicating that the path by which E affects D by affecting sr (i.e., t1→E→sr→D) and the path by which sr directly affects D are more critical to the somatosensory index D.

[0022] In one embodiment, the specific steps of using the Bayesian network to update parameter weights in real time are: Bayesian network structure learning: Based on the causal relationship between parameters, the Bayesian network structure is preliminarily constructed. The K2 algorithm is used to optimize the preliminarily constructed network structure to find the network structure that can most accurately describe the dependency relationship between parameters. Parameter learning: Based on the determined network structure, the conditional probability distribution of each node in the network is learned by maximum likelihood estimation using the collected data; Parameter weight calculation and screening: Calculate the influence weight of each parameter on the somatosensory index D, measure the correlation between the parameter and the somatosensory index D by calculating the mutual information, set a threshold, and only retain the parameters with significant impact on the somatosensory index D; Model update and optimization: Regularly use new data to re-learn parameters and recalculate weights, update the Bayesian network model, and adapt to environmental changes and dynamic changes in user preferences.

[0023] In this example, a study is conducted on temperature and humidity regulation in an office environment. The goal is to find an optimal temperature and humidity combination to improve employee comfort. In this scenario, the environmental parameters of interest include temperature (T), humidity (H), surface humidity (sr), and body temperature (Body_Temp). The somatosensory index (D) serves as a comprehensive indicator for measuring employee comfort.

[0024] Preliminary network structure construction: Based on the known causal relationship between parameters (temperature affects humidity, humidity and temperature jointly affect body surface humidity, etc.), a Bayesian network structure was preliminarily constructed. In this structure, temperature (T), humidity (H), body surface humidity (sr), and body temperature (Body_Temp) serve as input nodes, and somatosensory index (D) serves as output node; Use the K2 algorithm to optimize the network structure: To find the network structure that most accurately describes the dependency relationship between parameters, the K2 algorithm was used to optimize the initially constructed network structure. Through the K2 algorithm, a more refined Bayesian network structure that conforms to the data dependency relationship was obtained. Parameter learning: Based on the determined network structure, parameter learning was performed using collected data (including measurements of temperature, humidity, surface humidity, body temperature, and somatosensory index). The maximum likelihood estimation method was used to learn the conditional probability distribution of each node in the network, and the conditional probability of each node under a given parent node state was obtained. Parameter weight calculation and screening: Calculate the influence weight of each parameter on the somatosensory index D, and measure the correlation between the parameter and the somatosensory index D by calculating the mutual information. The greater the mutual information, the higher the correlation between the parameter and the somatosensory index D. Set a threshold to filter parameters: Set a threshold. Only when the mutual information between the parameter and the somatosensory index D is greater than this threshold, we consider that the parameter has a significant impact on the somatosensory index D and retain it in the model. Model update and optimization: Regularly use new data to re-learn parameters and recalculate weights to update the Bayesian network model.

[0025] In one embodiment, the specific steps of generating a local eddy current model using a lightweight CFD simulation tool are as follows: Establish an initial model: Determine the spatial scope of the simulation and set the boundary conditions of the space. For example, set the no-slip boundary condition for walls and the airflow exchange conditions at doors and windows based on their opening and closing conditions. Add obstacles (such as furniture) and a geometric model of the human body to the simulation area. The human body can be a simplified 3D model, considering the impact of the main contours and posture of the human body on airflow. Meshing: Use mesh generation tools to divide the simulation area into small grid cells. Local mesh refinement is performed near obstacles and human shapes. For example, near the human body surface, the mesh size can be set to a smaller value, such as 0.05m, while in areas away from the human body, the mesh size can be appropriately increased, such as 0.1m. This ensures that the mesh quality meets the simulation requirements and avoids the occurrence of deformed meshes that affect the accuracy of the simulation results.

[0026] Set physical parameters: Airflow parameters: Set physical parameters such as air density and viscosity, as well as initial conditions such as air velocity and temperature at the inlet; Humidity parameters: Set initial humidity distribution and set the initial humidity value in the room according to actual conditions; Run lightweight CFD simulation: Use lightweight CFD simulation tools, such as a simplified version of OpenFOAM, to perform simulation calculations and solve the Navier-Stokes equations (NS equations) and energy equations to obtain the velocity field, pressure field, and temperature field distribution of the airflow. The NS equations are expressed in rectangular coordinates as ; Among them, u, v, w are the velocity components in the x, y, and z directions respectively, and p is the pressure. is the air density, ν is the kinematic viscosity of air; Analyze simulation results: Extract the distribution information of local eddies from the simulation results, including the location, intensity, and direction of the eddies. The location and intensity of the eddies are determined by analyzing the curl of the velocity field. The curl formula is: ,in, is the curl vector, are the unit vectors in the x, y, and z directions respectively. Observe the changes in airflow around obstacles and human shapes, and analyze their effects on airflow speed, direction, and vortex formation.

[0027] In one embodiment, the relationship between humidity, air flow velocity and vortex intensity is analyzed based on the CFD simulation results to establish the following relationship: ,in, is the corrected humidity value, is the initial humidity value, is the air flow velocity, is the eddy current intensity, and is the correction factor.

[0028] In one embodiment, the specific steps of establishing a sweat evaporation dynamics model and calculating body surface humidity are: Parameter determination: Determine skin impedance, which is acquired in real time by the impedance sensor on the flexible electronic skin patch; determine airflow velocity, which is collected by the microenvironment sensor array; Model establishment: Using the sweat evaporation dynamics model formula ,in, is the body surface humidity, is the temperature difference between core and skin, is physiological data, expressed as skin impedance, is the air flow velocity, and are individual calibration parameters.

[0029] Parameter calibration: Through experiments, we collect the skin impedance, air flow velocity, and body surface humidity data of individuals under different environmental conditions. and Calibration is performed to measure the difference between the model prediction value and the actual measurement value through the model mean square error (MSE), and the model parameters are adjusted through the genetic algorithm to minimize the loss function; Calculation of body surface humidity: Substitute the real-time skin impedance and airflow velocity into the calibrated model formula to calculate the body surface humidity sr.

[0030] In one embodiment, the specific steps of the inner-loop dual-loop reinforcement learning (RL) adjustment are: Define the state space: The state space S includes the current ambient temperature, humidity, body surface humidity, body temperature, and somatosensory index; Define the action space: The action space A includes the temperature adjustment amount Δt1 and the humidity adjustment amount ΔE; Define the reward function: The reward function R is based on the user's real-time comfort score S: ; in, Indicates the status of the comfort score, , is the user's current comfort rating, is the user's last comfort rating; Selection strategy: Use Q-learning to select action a based on the current state s to maximize the long-term reward. The Q-learning update formula is expressed as ; in, is the learning rate, is the discount factor, s is the current state, a is the current action, r is the reward, and s' is the next state; Execute actions and update states: After executing action a, observe the new state s' and reward r, and update the Q value or policy parameters.

[0031] In this example, assume a summer office environment with an initial temperature of 30°C and humidity of 70%. The user's body surface humidity is 50%, their body temperature is 36.8°C, and their comfort index D = 4 (on a scale of 1-5, with 5 being the most comfortable). User feedback: The user submitted a current comfort rating of 3 (moderate discomfort) via the mobile app. Goal: Through inner-loop RL adjustments, increase the comfort rating to above 4.

[0032] Step 1: Define the state space and action space Current status(s): ; Action Space (A): Temperature adjustment ; Humidity adjustment ; Step 2: Define the reward function The reward function changes based on the user's real-time comfort score: ; Initial rating: (The user's last rating was 3 points); Target: If the action is performed , then the reward R=4-3=+1; if , then R=-1.

[0033] Step 3: Q-learning strategy selects actions Assume that the initial Q table is empty and the system randomly selects actions: Action selection: Δt1=−1°C (cooling by 1°C), ΔE=−5% (dehumidification by 5%); Environmental parameters after the action: T = 29 ° C, RH = 65%, sr = 45% (due to reduced humidity, evaporation from the body surface is accelerated), tr = 36.7 ° C, D = 5; User feedback: The comfort score increased to 4 points, and the reward was R = 4 − 3 = + 1.

[0034] Step 4: Q-value update Use Q-learning to update the formula: Parameter settings: learning rate α=0.1, discount factor γ=0.9; Initial Q value: Q(s,a)=0 (first execution); Next state (s'): S' = [29℃, 65%, 45%, 36.7℃, 5].

[0035] Calculate the Q value: Q(s,a)←0+0.1×[1+0.9×0−0]=0.1.

[0036] Step 5: Continuous Learning and Optimization Second round of adjustments: The current state S' = [29℃, 65%, 45%, 36.7℃, 5], and the user rating remains 4 points.

[0037] The system selects a new action (such as maintaining the current settings).

[0038] If the user rating does not change (R=0), the Q value update weight decreases.

[0039] The third round of adjustments: Assume that the ambient temperature rises to 30°C due to external factors and the user rating drops to 3 points.

[0040] The system chooses a new action (such as turning on a fan to increase the wind speed) and is rewarded R=−1.

[0041] After the Q value is updated, the system avoids performing the same action in similar states.

[0042] In one embodiment, the outer-loop dual-loop reinforcement learning (RL) prediction steps are as follows: Collect historical data: Collect user temperature and humidity preference data in different seasons and circadian rhythms; Define the state space: The state space includes time (season, day and night), historical temperature and humidity setting values, and user feedback; Define the action space: The action space is the long-term temperature and humidity preference prediction value; Define the reward function: The reward function is based on how well the predicted preferences match the actual preferences; Train the prediction model: Use historical data to train the RL model to predict long-term user preferences.

[0043] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for regulating environmental temperature and humidity based on body comfort, characterized in that: The following steps are involved: Step 1: Environmental parameter and physiological data collection: A microenvironmental sensor array is used to collect 3D distribution data of temperature, humidity, and airflow velocity in a specified area in real time. The sensor data is processed by the edge computing layer to generate a thermal map of the environmental parameters. A flexible electronic skin patch is used to obtain skin surface electrophysiological data in real time. The patch data is also processed by the edge computing layer to extract key parameters related to sweat secretion and skin condition, providing input for the sweat evaporation dynamics model. Step 2: Dynamic Causal Graph Model Analysis: Analyze the causal relationship between ambient temperature, ambient humidity, body surface humidity, and body temperature parameters, construct a dynamic causal graph model, identify key paths, and determine the degree of influence of each parameter on the somatosensory index. Use a Bayesian network to update parameter weights in real time, retaining only parameters with a significant impact on the somatosensory index. Step 3: CFD lightweight simulation: Use lightweight CFD simulation tools to generate a local eddy current model. Based on the CFD simulation results, dynamically modify the calculation formula for ambient humidity. Step 4: Calculation of the sweat evaporation dynamics model: Establish a sweat evaporation dynamics model, update the ambient humidity boundary conditions based on the corrected ambient humidity, calculate the body surface humidity, and align the thermal map of the environmental parameters with the physiological data in time and space to form a multimodal input vector, which serves as the initial and boundary conditions of the sweat evaporation dynamics model. Step 5: Dual-loop reinforcement learning framework adjustment: Inner-loop dual-loop reinforcement learning adjustment: Uses the user's real-time comfort score as a reward to dynamically adjust the temperature and humidity set points. When the user's feedback shows a decrease in comfort, dual-loop reinforcement learning automatically adjusts the ambient temperature. Outer-loop dual-loop reinforcement learning prediction: Combines historical data to predict the user's long-term preferences. Step 6: Adversarial safety constraint assurance: Control the temperature and humidity adjustment within the medical safety range. Introduce a penalty term in the double-loop reinforcement learning. When the adjustment instruction exceeds the safety range, a negative reward is given. Step 7: Generate and execute adjustment instructions: Generate adjustment instructions based on user physiological data and environmental parameters, and execute the adjustment instructions through the air conditioning controller and humidifier / dehumidifier.

2. The method for adjusting environmental temperature and humidity based on body comfort according to claim 1, characterized in that: The specific steps of constructing the dynamic causal graph model are: Parameter causal relationship analysis: Granger causality test was used to determine the causal relationship between ambient temperature, humidity, surface humidity, body temperature and somatosensory index; Construct a dynamic causal graph model: Based on the causal relationships obtained from the above analysis, use graphical tools to construct a dynamic causal graph model. In the graph, nodes represent each parameter, and directed edges represent the causal relationships between parameters. Identify key paths: Determine the influence weight of each path on the somatosensory index D. Use regression analysis method, take the somatosensory index as the dependent variable, and take ambient temperature, humidity, body surface humidity, and body temperature as independent variables to establish a regression model. The coefficient of the regression model is used to determine the influence degree of each path.

3. The method for adjusting environmental temperature and humidity based on body comfort according to claim 2, characterized in that: The specific steps of using the Bayesian network to update parameter weights in real time are: Bayesian network structure learning: Based on the causal relationship between parameters, the Bayesian network structure is preliminarily constructed. The K2 algorithm is used to optimize the preliminarily constructed network structure to find the network structure that can most accurately describe the dependency relationship between parameters. Parameter learning: Based on the determined network structure, the conditional probability distribution of each node in the network is learned by maximum likelihood estimation using the collected data; Parameter weight calculation and screening: Calculate the influence weight of each parameter on the somatosensory index D, measure the correlation between the parameter and the somatosensory index D by calculating the mutual information, set a threshold, and only retain the parameters with significant impact on the somatosensory index D; Model update and optimization: Regularly use new data to re-learn parameters and recalculate weights, update the Bayesian network model, and adapt to environmental changes and dynamic changes in user preferences.

4. The method for adjusting environmental temperature and humidity based on body comfort according to claim 1, characterized in that: The specific steps of generating a local eddy current model using a lightweight CFD simulation tool are as follows: Establish the initial model: determine the spatial scope of the simulation, set the boundary conditions of the space, and add obstacles and human body geometry models to the simulation area; Meshing: Use mesh generation tools to divide the simulation area into small mesh cells, and perform local mesh refinement near obstacles and human shapes; Set physical parameters: airflow parameters: set the physical parameters of the air and the initial conditions at the inlet; humidity parameters: set the initial humidity distribution; Run lightweight CFD simulation: Use lightweight CFD simulation tools to solve the Navier-Stokes equations and energy equations to obtain the velocity field, pressure field, and temperature field distribution of the airflow. The NS equations are expressed in rectangular coordinates as follows: ； Among them, u, v, w are the velocity components in the x, y, and z directions respectively, and p is the pressure. is the air density, ν is the kinematic viscosity of air; Analyze simulation results: Extract the distribution information of local eddies from the simulation results, including the location, intensity, and direction of the eddies. The location and intensity of the eddies are determined by analyzing the curl of the velocity field. The curl formula is: ,in, is the curl vector, are the unit vectors in the x, y, and z directions respectively. Observe the changes in airflow around obstacles and human shapes, and analyze their effects on airflow speed, direction, and vortex formation.

5. The method for regulating environmental temperature and humidity based on body comfort according to claim 4, characterized in that: Based on the CFD simulation results, the relationship between ambient humidity, air flow velocity and vortex intensity is analyzed, and the following relationship is established ,in, is the corrected ambient humidity value, is the initial humidity value, is the air flow velocity, is the eddy current intensity, and is the correction factor.

6. The method for adjusting environmental temperature and humidity based on body comfort according to claim 1, characterized in that: The specific steps of establishing the sweat evaporation dynamics model and calculating the body surface humidity are as follows: Parameter determination: Determine skin impedance, which is acquired in real time by the impedance sensor on the flexible electronic skin patch; determine airflow velocity, which is collected by the microenvironment sensor array; Model establishment: Using the sweat evaporation dynamics model formula ,in, is the body surface humidity, is the temperature difference between core and skin, is physiological data, expressed as skin impedance, is the air flow velocity, and Calibrate parameters for individuals; Parameter calibration: Through experiments, we collect the skin impedance, air flow velocity, and body surface humidity data of individuals under different environmental conditions. and Calibration is performed to measure the difference between the model prediction value and the actual measurement value through the model mean square error, and the model parameters are adjusted through the genetic algorithm to minimize the loss function; Calculation of body surface humidity: Substitute the real-time skin impedance and airflow velocity into the calibrated model formula to calculate the body surface humidity sr.

7. The method for regulating environmental temperature and humidity based on body comfort according to claim 1, characterized in that: The specific steps of the inner-loop dual-loop reinforcement learning adjustment are: Define the state space: The state space S includes the current ambient temperature, humidity, body surface humidity, body temperature, and somatosensory index; Define the action space: The action space A includes the temperature adjustment amount Δt1 and the humidity adjustment amount ΔE; Define the reward function: The reward function R is based on the user's real-time comfort score S: ； in, Indicates the status of the comfort score, , is the user's current comfort rating, is the user's last comfort rating; Selection strategy: Use Q-learning to select action a based on the current state s to maximize the long-term reward. The Q-learning update formula is expressed as ； in, is the learning rate, is the discount factor, s is the current state, a is the current action, r is the reward, and s' is the next state; Execute actions and update states: After executing action a, observe the new state s' and reward r, and update the Q value or policy parameters.

8. The method for adjusting environmental temperature and humidity based on body comfort according to claim 1, characterized in that: The specific steps of the outer-loop dual-loop reinforcement learning prediction are: Collect historical data: Collect user temperature and humidity preference data in different seasons and circadian rhythms; Define the state space: The state space includes time, historical temperature and humidity settings, and user feedback; Define the action space: The action space is the long-term temperature and humidity preference prediction value; Define the reward function: The reward function is based on how well the predicted preferences match the actual preferences; Train the prediction model: Use historical data to train the RL model to predict long-term user preferences.

Images