A heating ventilation air conditioning intelligent regulation and control method based on user portrait

Abstract

The application discloses a kind of based on user portrait's heating ventilation air conditioner intelligent regulation and control method.The method includes the following steps: 1) obtain the video image of user;2) training classification based on CNN network, obtain the dress and posture of user;3) obtain user distance, and arrange temperature sensor, construct a BP neural network to predict the air temperature at user distance;4) user distance and the air temperature at user distance are input into the evaluation model based on DQN network structure for temperature prediction;5) in combination with the dress and posture of user, the air supply mode of all air conditioners is evaluated, and the optimal air supply mode is obtained;6) the optimal air supply mode is input into heating ventilation air conditioner.A kind of based on user portrait's heating ventilation air conditioner intelligent regulation and control method in the application is based on comfort zone temperature, in combination with deep learning algorithm and temperature prediction model, predicts the air temperature of each user under different heating ventilation air conditioner settings, and finally selects the best heating ventilation air conditioner setting according to scoring model.

Description

Technical Field

[0001] This invention relates to the field of home appliances, and in particular to a method for intelligent control of HVAC systems based on user profiles. Background Technology

[0002] Individual thermal comfort is a crucial factor influencing overall user satisfaction with the indoor environment. In economically developed cities, people spend over 90% of their time indoors. Therefore, the indoor environment has a profound impact on people's physical and mental health.

[0003] Thermal comfort is defined as a psychological state expressing satisfaction with one's thermal environment. People's thermal comfort depends not only on environmental factors such as temperature and humidity, but also on physiological factors (such as gender and heart rate), psychological factors (such as stress and mindset), and behavioral factors (such as activity level and clothing). Therefore, thermal comfort varies from person to person. Furthermore, due to differences in climate, even within China, there are significant differences in thermal comfort between people in the south and the north.

[0004] In most office buildings, administrators typically select central HVAC setpoints based on industry guidelines (ASHRAE). However, this traditional strategy of setting HVAC to fixed values ​​is unlikely to simultaneously meet the thermal comfort needs of multiple users for several reasons. First, indoor air temperature is uneven, and a fixed-value HVAC supply strategy cannot guarantee that the air temperature at different locations within a room remains at the set temperature. Furthermore, people sitting near air vents or in direct sunlight may experience different levels of heat than others. Finally, using indoor temperature alone as the sole indicator of thermal comfort is insufficient to reflect users' perceived thermal comfort. Moreover, several studies have shown that even when room temperature is set according to recommended indoor conditions, user dissatisfaction with indoor temperature environments remains high, reflecting a discrepancy between actual and predicted thermal comfort levels.

[0005] The Predictive Average Votes (PMV) model is the most commonly used method for evaluating thermal comfort. The PMV model considers four environmental factors: air temperature, relative humidity, air velocity, and mean radiant temperature, as well as two human factors: human metabolic rate and clothing grade, and predicts the user's average thermal sensation on a 7-point scale (from cold to hot).

[0006] Currently, research on HVAC air supply strategies based on thermal comfort mainly falls into two categories: one is based on human feedback for adjustment, and the other is based on human biological signals for adjustment.

[0007] Methods based on human feedback for temperature regulation can be broadly categorized into two types: PMV-based methods and non-PMV-based methods. In PMV-based methods, researchers collect user feedback on thermal comfort using sensors and mobile applications. The decision-making module then adjusts the HVAC setpoints in real time based on the PMV model or overall user feedback. However, these PMV-based methods have limitations. For example, they require assumptions of a steady-state air conditioning environment, cannot predict transient responses, and require multiple parameters to estimate the PMV model (such as mean radiant temperature and metabolic rate). In practical applications, these parameters may deviate significantly from actual conditions.

[0008] In non-PMV-based methods, researchers typically use data collected from the indoor environment to simulate user thermal comfort, and then adjust the HVAC setpoints in real time based on industry guidelines (ASHRAE) recommended setpoints and the simulated user thermal comfort. However, this non-PMV method based on collected user thermal preferences (from cold to hot) also has certain limitations, such as a lack of human body data (e.g., skin temperature, activity level), and the potential time lag in user feedback-based methods, which may not meet the thermal comfort needs of all users.

[0009] In methods based on human biosignals, researchers assess the level of human thermal comfort under laboratory conditions using human biosignals (such as skin temperature and heart rate). However, these methods are all conducted under laboratory conditions, and therefore, in practical use, they are limited by environmental and human factors (such as humidity, airflow, and clothing), and their improvement on users' thermal comfort is not ideal. Summary of the Invention

[0010] To address the above problems, this invention provides a method for intelligent control of HVAC systems based on user profiles.

[0011] According to one aspect of the present invention, a method for intelligent control of HVAC based on user profiles is provided, comprising the following steps:

[0012] 1) Use a stereo camera to obtain video images of the user;

[0013] 2) Training and classification based on a CNN network to obtain the user's clothing and posture;

[0014] 3) Obtain the user distance and deploy temperature sensors to construct a BP neural network to predict the air temperature at the user distance;

[0015] 4) Input the user distance and the air temperature at the user distance into the evaluation model for temperature prediction based on the DQN network structure;

[0016] 5) Evaluate all air supply modes of the air conditioner based on the user's clothing and posture to obtain the optimal air supply mode;

[0017] 6) Input the optimal air supply mode into the HVAC system.

[0018] In some implementations, in step 1), the calibrated binocular camera is mounted in a position aligned with the air conditioner horizontally to obtain a video image. The advantage of this is that it describes the mounting position of the binocular camera.

[0019] In some implementations, in step 2), a training set of clothing or pose information is selected, and a CNN network is used for training and classification. Then, single-frame images are acquired from a binocular camera, and the images are segmented for different users. The segmented image of each user is then fed into the trained network for recognition, thereby obtaining the user's clothing and pose. Its advantage lies in describing a method for obtaining the user's clothing and pose.

[0020] In some implementations, in step 3), the distance between the user and the camera, and the horizontal distance between the binocular camera and the HVAC system are obtained, thereby determining the user distance. The advantage of this approach is that it describes how the user distance is obtained.

[0021] In some implementations, in step 3), the input layer of the BP neural network contains 6 neurons, the intermediate layers are divided into 3 layers containing 100, 200, and 100 neurons respectively, and the output layer contains 1 neuron. Its advantage lies in describing the structure of the BP neural network.

[0022] In some implementations, in step 4), the input layer of the temperature prediction network evaluation model based on the DQN network structure contains 2 neurons, the middle layer is divided into 3 layers containing 200, 300, and 500 neurons respectively, and the output layer contains 270 neurons. Its advantage lies in describing the structure of the temperature prediction network evaluation model based on the DQN network structure.

[0023] In some implementations, step 4) of the temperature prediction algorithm based on the DQN network structure includes the following steps:

[0024] A) Initialize the playback memory unit;

[0025] B) Initialize the value network with random weights;

[0026] C) Initialize the target value network using the target weights;

[0027] D) Record the collected information into the playback memory unit;

[0028] E) Randomly select training samples in small batches from the playback memory units;

[0029] F) Train a value network with the aforementioned random weights based on the error function;

[0030] G) Every N time steps, copy the parameters of the value network to the target value network.

[0031] Its advantage lies in describing the specific real-time steps of the algorithm for temperature prediction based on the DQN network structure.

[0032] In some implementations, during steps 2) to 4), when there are multiple users, the obtained user clothing and posture, user distance, and air temperature at the user distance are all numbered. The advantage of this is that numbering these parameters distinguishes different users and facilitates subsequent calculations.

[0033] In some implementations, the process of evaluating the air supply mode in step 5) includes the following steps:

[0034] a) Set up a humidity sensor for measurement;

[0035] b) Obtain the thermal resistance value of the user's clothing;

[0036] c) Based on the humidity sensor values ​​and the different thermal resistance values ​​of clothing given by the industry standard ASHRAE, obtain the recommended operating temperature range for clothing with thermal resistance values ​​of 1 col and 0.5 col, and calculate the boundary of the comfortable recommended operating temperature range;

[0037] d) Based on the user's posture, first obtain the user's average radiant temperature, then combine it with the boundary of the recommended comfortable operating temperature range to obtain the user's comfortable air temperature range;

[0038] e) Use the midpoint of the comfortable air temperature range as the optimal operating temperature;

[0039] f) Users award rewards based on the optimal operating temperature, and an evaluation function for the total reward is constructed;

[0040] g) Select the air supply mode with the highest total reward as the optimal air supply mode.

[0041] Its advantage lies in describing the specific steps for evaluating air supply patterns.

[0042] In some implementations, in step f), the evaluation function for the total reward is:

[0043] ,

[0044] Where n is the number of users, R i The reward for one of the users is as follows:

[0045] R i = .

[0046] Its advantage lies in describing the content of the evaluation function of the total reward. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating a user profile-based intelligent control method for HVAC systems according to one embodiment of the present invention.

[0048] Figure 2 for Figure 1 The diagram shows the structure of an evaluation model for temperature prediction based on a DQN network.

[0049] Figure 3 for Figure 1 The diagram shows different thermal resistance values ​​for clothing as specified in the industry standard (ASHRAE).

[0050] Figure 4 for Figure 1 The diagram shows the experimental effect of a fixed HVAC system.

[0051] Figure 5 for Figure 1 The diagram shows the experimental effect of intelligent control of HVAC. Detailed Implementation

[0052] The present invention will now be described in further detail with reference to the accompanying drawings.

[0053] like Figure 1 As shown in the figure, this is the workflow of the intelligent control method for HVAC based on user profiles. The method includes several main steps, which are described below.

[0054] The first step is to obtain the user's video image.

[0055] In this step, a calibrated binocular camera is installed in the user's HVAC room to obtain video images. The binocular camera should be installed in a horizontal line with the air conditioner. The selected HVAC settings are: temperature range [19°C~28°C], airflow set to [1, 2, 3], three lateral airflow directions [left, center, right], and three longitudinal airflow directions [up, center, down], totaling 270 HVAC settings actions.

[0056] The second step is to obtain the user's clothing and posture.

[0057] In this step, a CNN network can be used for training and classification to obtain information such as the user's clothing and posture.

[0058] The method for obtaining a user's clothing information is as follows: DeepFashion, a large-scale dataset released by the Chinese University of Hong Kong, is used as the training set. The dataset is trained and classified based on a CNN network. Then, single-frame images are acquired from a stereo camera, and the images are segmented for different users. The image of each segmented user is then fed into the trained network for recognition, thereby obtaining the user's clothing information.

[0059] In this paper, 10 types of user clothing are identified, including: short-sleeved T-shirt, long skirt, jeans, short skirt, long-sleeved T-shirt, etc.

[0060] The method for obtaining a user's pose is as follows: Select images of standing and sitting human poses from the publicly available human pose dataset (MPII Human Pose dataset) as the training set, and train and classify them based on a CNN network. Then, acquire single-frame images from a binocular camera, segment the images for different users, and put the segmented images of each user into the trained network for recognition, thereby obtaining the user's pose.

[0061] For user clothing and posture information that cannot be accurately obtained, default values ​​are used to fill in the information. The default value for thermal resistance of the user clothing is 0.5clo, and the default value for user posture is sitting.

[0062] In addition, considering that there are usually multiple users indoors during the experiment, the clothing and posture of each user will be numbered in this step.

[0063] The third step is to obtain the user's distance and predict the air temperature at that distance.

[0064] In this step, based on the principle of binocular ranging, the distance d between the user and the camera can be obtained. c To obtain a better field-of-view image, the binocular camera is installed at a certain downward angle β, thus the horizontal distance d between the user and the binocular camera can be obtained. cl for:

[0065]

[0066] Since the binocular camera is installed on the same horizontal line as the air conditioner, the horizontal distance d between the binocular camera and the HVAC system is then measured. ct Therefore, the horizontal distance between the user and the HVAC system, i.e., the user distance d, can be derived as:

[0067] .

[0068] Then, multiple temperature sensors can be installed indoors to measure the temperature. These sensors are typically positioned at different distances from the HVAC system, with the sensors 1 meter above the ground. The HVAC system is set to refresh the air supply every hour, while each temperature sensor collects data every 10 minutes. The collected data is then fed into a neural network for training.

[0069] Analysis of the data collected by all temperature sensors reveals that the temperatures collected by sensors equidistant from the HVAC center at the same time differ by less than 0.1°C. Therefore, the distance to the HVAC center can be used instead of the traditional planar position representation; that is, a single parameter, distance d, can be used to represent the position information using a planar coordinate system. This not only reduces the difficulty of representing the user's real-time location but also reduces the number of parameters, thereby reducing model training time.

[0070] Next, a backpropagation (BP) neural network is trained using data collected by a temperature sensor to predict the air temperature at a distance *d* from the user. The BP neural network has the following structure: an input layer containing 6 neurons, corresponding to the user's distance *d*, the air temperature *t* at distance *d*, the air conditioning setting, and the fan speed, lateral airflow, and longitudinal airflow (the fan speed, lateral airflow, and longitudinal airflow settings are digitized using numbers 1, 2, and 3); three middle layers containing 100, 200, and 100 neurons respectively; and an output layer containing 1 neuron, corresponding to the air temperature at distance *d* after 10 minutes. .

[0071] In addition, this step will also number the distances to each user and the air temperature at each user distance.

[0072] The fourth step involves inputting the user distance and the air temperature at that distance into an evaluation model for temperature prediction based on a reinforcement learning (DQN) network structure.

[0073] In this step, the evaluation model of the temperature prediction network based on DQN is set as follows: the input layer contains 2 neurons, corresponding to the user distance d and the air temperature t at the distance d; the middle layer has 3 layers, each containing 200, 300, and 500 neurons; the output layer has 270 neurons, corresponding to 270 HVAC setting actions, and the prediction accuracy is 0.1.

[0074] like Figure 2 As shown in the figure, this is the structure of an evaluation model for temperature prediction based on a DQN network. Here, 's' represents the collected information, including the user's distance 'd' and the air temperature 't' at distance 'd'. 'a' represents the HVAC setting action. This indicates the air temperature measured 10 minutes later at a distance d under the corresponding HVAC setting. The temperature predicted by the target value network is also the output value of the trained network. The temperature is the current value predicted by the network. The error function is... .

[0075] The temperature prediction algorithm based on the DQN network structure includes several steps, A) to G), which are described below.

[0076] A) Initialize the playback memory unit.

[0077] B) Using random weights Initialize the value network.

[0078] C) Using target weights ( Initialize the target value network.

[0079] D) The collected information ( Recorded in the playback memory unit.

[0080] E) Randomly select training samples in small batches from the playback memory units.

[0081] F) Train a value network with the above random weights based on the error function.

[0082] G) Every N time steps, copy the parameters of the value network to the target value network. .

[0083] The fifth step is to evaluate all the air supply modes of the air conditioner based on the user's clothing and posture, and obtain the optimal air supply mode.

[0084] In this step, several concepts need to be clarified, including the dry-bulb temperature T. db Mean radiation temperature T MR Operating temperature T op And the thermal resistance value Icl of clothing.

[0085] Dry bulb temperature T db Thermal comfort refers to the temperature of the indoor air surrounding a person, measured from a dry-bulb thermometer exposed to air but not directly exposed to sunlight. Generally, most industry guidelines recommend a comfortable temperature range of 18 to 23 degrees Celsius, with a temperature difference not exceeding 1 degree Celsius. For sedentary or near-sedentary physical activity levels, i.e., typical office activities, the recommended comfortable temperature range for optimizing indoor thermal comfort is 19 to 28 degrees Celsius, with a temperature difference not exceeding 1 degree Celsius. However, the required comfortable temperature range may vary depending on clothing worn in different seasons.

[0086] Mean radiation temperature T MRThis refers to the average temperature of the radiation emitted by the surrounding surfaces of the environment to the human body. All surfaces in the room are considered to be uniformly black. The value can be determined by the surface temperature and the angular coefficient of the relationship between the person and the surface, or by measuring with a black sphere thermometer. The average radiant temperature can also be calculated from the planar radiant temperature of the human body in six directions (up, down, left, back, front, and back) and the projected area of ​​the human body in the six directions.

[0087] For a standing person, the average radiant temperature can be estimated using the following formula:

[0088] (1)

[0089] For a seated person, the average radiant temperature can be estimated using the following formula:

[0090] (2)

[0091] Operating temperature T op Derived from air temperature, mean radiant temperature, and air velocity, this represents the thermal effect of both air temperature and mean radiant temperature on the human body. It can be considered a weighted average of indoor air temperature and mean radiant temperature under a certain coefficient, reflecting the temperature at which the environment exerts a thermal effect on the human body. The effective temperature T is... op It is defined as the temperature at which the heat exchanged between human bodies through radiation and convection in a space with a uniform black inner surface is equal to the dry heat loss of the human body in a real non-uniform environment.

[0092] Operating temperature T op The simplified expression for is the following formula:

[0093] (3)

[0094] Clothing thermal resistance value I cl Thermal resistance refers to the ratio of the temperature difference between the two ends of clothing per unit area to the power of the heat source when heat is transferred. It represents the insulation capacity of a person's clothing, and its unit is clo (1 clo = 155 m²℃W). The overall thermal resistance of a person's clothing can be estimated from the thermal resistance of individual garments using a summation formula. The expression for the thermal resistance of clothing can be given by the following formula:

[0095] (4)

[0096] in It is the effective thermal resistance value of garment i.

[0097] The thermal resistance values ​​of some garments are shown in Table 1:

[0098] Table 1 Thermal resistance values ​​of clothing

[0099]

[0100] like Figure 3 As shown in the figure, the thermal resistance values ​​for different garments are given by the industry standard (ASHRAE). The airflow modes are evaluated using the industry standard (ASHRAE). Let T0 be the recommended operating temperature under the humidity conditions, which is the effective temperature. Then, the following formula applies:

[0101] T 0= T op (5)

[0102] Meanwhile, the dry-bulb temperature is taken as the comfortable air temperature T. air Then we have the following formula:

[0103] T air= T db (6)

[0104] The evaluation process for air supply modes mainly includes several steps, a) to g), which are described below.

[0105] a) Install a humidity sensor to measure the indoor humidity.

[0106] b) According to formula (4), the thermal resistance value I of the user's clothing is obtained. cl .

[0107] c) Based on the humidity sensor value, combined with Figure 3 As shown, the recommended operating temperature ranges for clothing with thermal resistance values ​​of 1col and 0.5col are obtained, and the range of comfortable recommended operating temperatures is calculated.

[0108] Among them, the recommended operating temperature T0 range when the thermal resistance of the clothing is 1col is [T min,1.0clo T max,1.0clo The recommended operating temperature range T0 for clothing with a thermal resistance of 0.5col is [T]. min,0.5clo T max,0.5clo The recommended operating temperature range is [ ]. The method for calculating the two end boundaries of ] is as follows:

[0109] (7)

[0110] (8)

[0111] d) Based on the user's posture, first obtain the user's average radiant temperature T. MR Then, by combining the boundaries of the recommended operating temperature range, the user's comfortable air temperature range can be obtained.

[0112] Combining formulas (3), (4), and (5), we can obtain the following formula:

[0113] (9)

[0114] Substitute the recommended operating temperature range for comfort. ], then the corresponding comfortable air temperature range is [ Substituting these values ​​into formulas (7) and (8), we can obtain the specific values ​​for the comfortable air temperature range.

[0115] e) Use the midpoint of the comfortable air temperature range as the optimal operating temperature T*.

[0116] Studies have shown that within the recommended operating temperature range given by industry guidelines (ASHRAE), according to the ASHRAE thermal sensation scale (see Table 2), a person will feel a thermal sensation of +0.5 near the boundary of the warmer zone; and a person may feel a thermal sensation of -0.5 near the boundary of the colder zone. In the middle of this specified zone, a person wearing the specified clothing will have a very close to neutral sensation. Therefore, the median value of the resulting comfortable air temperature range is taken as the optimal operating temperature T*.

[0117] Table 2 ASHRAE Thermal Sensation Scale

[0118]

[0119] f) Each user using the air conditioner is given a reward based on the above-mentioned optimal operating temperature T*, and an evaluation function for the total reward is constructed.

[0120] The final total reward evaluation function is as follows:

[0121] ,

[0122] Where n is the number of users, R i The reward for one of the users is as follows:

[0123] R i = .

[0124] g) Evaluate all HVAC air supply modes using an evaluation model, and finally select the air supply mode with the highest total reward r as the optimal air supply mode.

[0125] To ensure system stability, the optimal air supply mode will only be sent to the HVAC system when the reward for the optimal air supply mode is 10% higher than the reward for the previous set air supply mode.

[0126] The sixth step is to input the optimal air supply mode into the HVAC system to control adjustments and ensure the user receives the most comfortable experience.

[0127] This method uses a computer instead of a decision-making module and controls a customized infrared transmitter via Python scripts, thereby controlling the HVAC system. This control method does not involve the details of the HVAC system and does not require modification of existing HVAC systems, significantly reducing the cost of developing and testing decision-making algorithms.

[0128] In addition, a comparative experiment can be conducted to test the effectiveness of the above-mentioned user profile-based intelligent control method for HVAC systems, as described below:

[0129] The experiment was conducted in a 70-square-meter conference room equipped with a separate HVAC unit, tables and chairs, and fitted with a binocular camera, temperature sensor, and humidity sensor. Data collection took place during the summer months of June and July, and all collected data came from this conference room.

[0130] Two comparative experiments were conducted in the conference room to verify the performance of the proposed method in improving user comfort. Experiment 1 involved a fixed HVAC setting of 26°C, airflow of 3, and both lateral and longitudinal airflow directions set to medium. Experiment 2 involved controlling the HVAC system using the intelligent control method described in this paper.

[0131] During the experiment, the initial indoor humidity (40%) and dry-bulb temperature (30°C) were kept consistent for both experiments, and each experiment lasted for one hour. A total of 20 participants were selected for the experiment, including 10 males and 10 females. All participants were free to move around indoors, and the meeting room doors and windows were kept closed during the experiment.

[0132] Each participant was asked to provide feedback every 10 minutes via mobile phone, indicating whether they felt comfortable, neutral, or uncomfortable. To ensure data stability, participants were prohibited from opening doors and windows.

[0133] User comfort is represented by two indicators:

[0134] (1) The comfort index based on the model, i.e. whether the air temperature is within the comfortable air temperature range obtained by the model;

[0135] (2) Based on participant comfort indexes, i.e., based on feedback from all participants. Participant comfort data were collected every 10 minutes. The power consumption of HVAC was recorded after the experiment.

[0136] like Figure 4-5 As shown, Figure 4 The results show the number of participants who felt comfortable in the fixed HVAC settings of Experiment 1. Figure 5The results show the number of participants who felt comfortable using the HVAC intelligent control method. In the two graphs, the horizontal axis represents time (min) and the vertical axis represents the number of people who felt comfortable. Line A shows the experimental effect of the model-based comfort index, and line B shows the experimental effect of the participant-based comfort index.

[0137] from Figure 4 and Figure 5 The comparison shows that with fixed HVAC settings, it is consistently difficult to meet the thermal comfort needs of all participants. However, the intelligent control method described in this paper can meet the thermal comfort needs of most participants in less time than with fixed HVAC, achieving this goal earlier (30 minutes for model-based comfort indices and 40 minutes for participant-based comfort indices). It also shows that, due to individual differences among participants, the thermal comfort level based on participant feedback is lower than that based on the model.

[0138] This paper provides a comprehensive review of HVAC air supply strategies based on thermal comfort and proposes an HVAC control framework capable of determining optimal air supply and temperature settings under different environmental and user behavior conditions. By acquiring user behavior information through deep learning techniques and combining it with deep reinforcement learning algorithms and relevant industry standards (ASHRAE), an intelligent HVAC air supply strategy is derived. Finally, experiments demonstrate that this additional user behavior information can significantly improve user thermal comfort, meeting the thermal comfort requirements of multi-user scenarios compared to traditional fixed HVAC settings.

[0139] The above descriptions are merely some embodiments of the present invention. Those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A method for intelligent control of HVAC based on user profiles, characterized in that: Includes the following steps 1) Use a stereo camera to obtain video images of the user; 2) Select a clothing or posture training set, train and classify based on a CNN network, then acquire single-frame images from a binocular camera, segment the images for different users, and put the segmented images of each user into the trained network for recognition, thereby obtaining the user's clothing and posture. 3) Obtain the user distance and deploy temperature sensors to construct a BP neural network to predict the air temperature at the user distance; 4) Input the user distance and the air temperature at the user distance into the evaluation model for temperature prediction based on the DQN network structure; 5) Evaluate all air supply modes of the air conditioner based on the user's clothing and posture to obtain the optimal air supply mode; 6) Input the optimal air supply mode into the HVAC system; In step 5), the process of evaluating the air supply mode includes the following steps: a) Set up a humidity sensor for measurement; b) Obtain the thermal resistance value of the user's clothing; c) Based on the humidity sensor values ​​and the different thermal resistance values ​​of clothing given by the industry standard ASHRAE, obtain the recommended operating temperature range for clothing with thermal resistance values ​​of 1 col and 0.5 col, and calculate the boundary of the comfortable recommended operating temperature range; d) Based on the user's posture, first obtain the user's average radiant temperature, then combine it with the boundary of the recommended comfortable operating temperature range to obtain the user's comfortable air temperature range; e) Use the midpoint of the comfortable air temperature range as the optimal operating temperature; f) Users award rewards based on the optimal operating temperature, and an evaluation function for the total reward is constructed; g) Select the air supply mode with the highest total reward as the optimal air supply mode.

2. The intelligent control method for HVAC based on user profiles according to claim 1, characterized in that: In step 1), the calibrated binocular camera is installed in a position that is on the same horizontal line as the air conditioner to obtain video images.

3. The intelligent control method for HVAC based on user profiles according to claim 1, characterized in that: In step 3), the distance between the user and the camera, and the horizontal distance between the binocular camera and the HVAC system are obtained, thus obtaining the user distance.

4. The intelligent control method for HVAC based on user profiles according to claim 3, characterized in that: In step 3), the input layer of the BP neural network contains 6 neurons, the intermediate layer is divided into 3 layers and contains 100, 200 and 100 neurons respectively, and the output layer contains 1 neuron.

5. A user profile-based intelligent control method for HVAC systems according to claim 1, characterized in that: In step 4), the input layer of the temperature prediction network evaluation model based on the DQN network structure contains 2 neurons, the middle layer is divided into 3 layers and contains 200, 300 and 500 neurons respectively, and the output layer contains 270 neurons.

6. The intelligent control method for HVAC based on user profiles according to claim 5, characterized in that: In step 4), the algorithm for temperature prediction based on the DQN network structure includes the following steps: A) Initialize the playback memory unit; B) Initialize the value network with random weights; C) Initialize the target value network using the target weights; D) Record the collected information into the playback memory unit; E) Randomly select training samples in small batches from the playback memory units; F) Train a value network with the aforementioned random weights based on the error function; G) Every N time steps, copy the parameters of the value network to the target value network.

7. A method for intelligent control of HVAC based on user profiles according to claim 1, characterized in that: In steps 2) to 4), when there are multiple users, the clothing and posture of the users, the distance between the users, and the air temperature at the distance between the users are all numbered.

8. The intelligent control method for HVAC based on user profiles according to claim 1, characterized in that: In step f), the evaluation function for the total reward is: ， Where n is the number of users, R i As a reward for one of the users, and there is R i = ； Among them, T air T* represents the optimal operating temperature for comfortable air temperature.

Images