Intelligent adjusting method for distributed air treatment equipment

Abstract

The invention provides an intelligent adjustment method for distributed air treatment equipment, and belongs to the technical field of air treatment equipment, and the method comprises the steps: inputting a personnel distribution image into an airflow field reconstruction model based on a graph neural network, outputting three-dimensional space flow field distribution data, calculating a personnel intensity index, and predicting an average thermal sensation index value; an air supply adjustment double-layer game model is established based on the data, the air supply angle and the air supply speed of an electric side supply nozzle and the air supply quantity of a lower air supply outlet are solved, an actuator is controlled to adjust air supply parameters, and the target air outlet temperature and the refrigerant flow of the heat exchanger are calculated through a thermodynamic entropy production minimization heat exchange optimization algorithm; and feedback correction is performed according to the actually predicted average thermal sensation index deviation to form closed-loop control, so that the technical problem that the distributed air treatment equipment is difficult to adjust the air supply parameters in real time according to the personnel distribution dynamic change so as to ensure the thermal comfort and the energy efficiency at the same time is solved.

Description

Technical Field

[0001] This invention belongs to the field of air handling equipment technology, and more specifically, relates to an intelligent adjustment method for distributed air handling equipment. Background Technology

[0002] In the control of distributed air handling equipment, traditional technologies mainly employ temperature feedback-based PID control or preset air supply modes to adjust equipment operating parameters. This involves collecting indoor temperature data using temperature sensors placed in fixed locations, calculating control parameters based on the deviation between the set temperature and the actual temperature, and driving actuators to adjust the air supply angle, speed, and volume. This method is widely used in air conditioning systems of large buildings such as office buildings and commercial complexes. However, traditional control methods suffer from response lag. When the distribution of people indoors changes rapidly, fixed-location sensors cannot promptly capture changes in the thermal comfort needs of local areas, leading to untimely adjustments to air supply parameters and resulting in overcooling or overheating in some areas. Furthermore, due to the lack of accurate prediction of the three-dimensional airflow field distribution, the air supply strategy often adopts a conservative high-volume operation mode to ensure coverage, causing unnecessary energy waste. The core problem that traditional technologies struggle to solve lies in the lack of intelligent adjustment methods capable of rapidly reconstructing the three-dimensional airflow field distribution and optimizing air supply parameters in real time based on dynamic changes in occupant distribution. This makes it impossible to achieve energy efficiency optimization while ensuring thermal comfort. Summary of the Invention

[0003] In view of this, the present invention provides an intelligent adjustment method for distributed air handling equipment, which can solve the technical problem in the prior art that distributed air handling equipment is unable to adjust the air supply parameters in real time according to the dynamic changes in the distribution of people in order to simultaneously ensure thermal comfort and energy efficiency.

[0004] This invention is implemented as follows: It provides an intelligent adjustment method for a distributed air handling unit. After the distributed air handling unit starts operating, a personnel detection camera collects images of the personnel distribution around the unit. A thermal comfort sensor collects data on dry-bulb temperature, relative humidity, mean radiant temperature, and air velocity in the personnel activity area. The personnel distribution images are input into an airflow field reconstruction model, which outputs three-dimensional spatial flow field distribution data. Based on the personnel distribution images, the personnel density index for each monitored area is calculated. Furthermore, based on the dry-bulb temperature, relative humidity, mean radiant temperature, and air velocity data, a predicted average thermal sensation index for the personnel activity area is calculated. Based on the personnel density index, the predicted average thermal sensation index, and the three-dimensional spatial flow field distribution data, a two-layer game model for air supply adjustment is established to solve for the air supply angle and air supply parameters of the electrically driven side nozzles. The system controls the air velocity and air volume supplied from the lower air outlet, adjusts the air supply angle and air velocity by controlling the electric side air supply nozzle actuator, and adjusts the air volume by controlling the air valve at the lower air outlet. Based on the temperature field distribution and the outdoor fresh air temperature and return air temperature, the system calculates the target outlet air temperature and refrigerant flow rate of the heat exchanger using a heat exchange optimization algorithm that minimizes thermodynamic entropy production. The system then controls the water flow regulating valve of the heat exchanger to adjust the flow rate. The system re-acquires images of personnel distribution, as well as data on dry-bulb temperature, relative humidity, mean radiant temperature, and air velocity to calculate the actual predicted average thermal sensation index. The system calculates the predicted average thermal sensation index deviation between the actual predicted average thermal sensation index and the target predicted average thermal sensation index. When the absolute value of the predicted average thermal sensation index deviation is greater than the comfort deviation threshold, the predicted average thermal sensation index deviation is used as a feedback correction parameter and returned to the air supply regulation two-layer game model for re-solving.

[0005] Among them, there are 4 personnel detection cameras, which are installed in the four directions of the distributed air handling equipment to collect information on the distribution of people within a 15m radius around the equipment from all directions.

[0006] The structure of the airflow field reconstruction model is a spatial flow field prediction network based on graph neural network. The three-dimensional space covered by the distributed air handling equipment is discretized into several spatial nodes. The airflow transmission relationship between adjacent spatial nodes is represented by directed edges to construct a spatial topology graph structure. The coordinates of the spatial nodes are embedded into the high-dimensional feature space through sine and cosine position encoding.

[0007] The spatial flow field prediction network consists of three graph convolutional layers and two fully connected layers. The first graph convolutional layer has an input dimension of 128 and an output dimension of 256. The second graph convolutional layer has an input dimension of 256 and an output dimension of 512. The third graph convolutional layer has an input dimension of 512 and an output dimension of 256. The two fully connected layers map the 256-dimensional features to 128 dimensions and the final output dimension, respectively. The output dimension includes four channels: temperature component and velocity components in three directions. Residual connections are used to add the input features to the outputs of each layer.

[0008] Among them, the sine and cosine position encoding maps the three-dimensional coordinates of spatial nodes into high-dimensional vectors through sine and cosine functions of different frequencies, enabling the spatial flow field prediction network to automatically perceive the spatial position relationship of spatial nodes when learning the airflow transmission law, thus avoiding feature confusion caused by spatial translation.

[0009] The steps for establishing the training dataset for the airflow field reconstruction model include building a distributed air handling equipment model with a scale of 1:5 in a laboratory environment, arranging 200 temperature sensors and 200 wind speed sensors around the distributed air handling equipment model to form a dense grid of measuring points, collecting time series data of temperature and wind speed at each measuring point under different air supply conditions, and collecting a total of 500 sets of flow field data under different air supply conditions by changing the air supply angle, air supply speed, personnel distribution pattern and obstacle position.

[0010] The training steps for the airflow field reconstruction model use mean squared error as the loss function, and the Adam optimizer is used to update the parameters. The initial learning rate is set to 0.001, and the learning rate is reduced to 0.5 times the original value every 30 training cycles. The batch size is set to 32, and the total number of training cycles is 200. Training is stopped early when the loss function value on the validation set no longer decreases after 20 consecutive training cycles.

[0011] The method for calculating the population density index is as follows: the population distribution image collected by the personnel detection camera is divided into several monitoring sub-regions. Each monitoring sub-region corresponds to a fan-shaped area around the device. The central angle of the fan-shaped area is 90 degrees and the radius is 15m. Human body detection is performed on the image of each monitoring sub-region to identify the number of people in the monitoring sub-region. The number of people is divided by the area of ​​the monitoring sub-region to obtain the population density index.

[0012] Among them, the predicted average thermal sensation index is the PMV value calculated according to the international standard ISO7730. The calculation formula takes into account six parameters: human metabolic rate, clothing thermal resistance, dry bulb temperature, average radiation temperature, air velocity and relative humidity. The dimensionless value characterizing human body's hot and cold sensation is obtained by solving the human body thermal balance equation.

[0013] The air supply regulation two-layer game model includes an upper-layer game model with the goal of maximizing comfort and a lower-layer game model with the goal of minimizing energy consumption. The objective function of the upper-layer game model is to maximize the weighted thermal comfort function, and the objective function of the lower-layer game model is to minimize the air supply energy consumption function. The two objective functions are related through a coupling term, which is the integral of the product of the temperature gradient and velocity gradient of each spatial node in the three-dimensional spatial flow field distribution data in space. The air supply regulation two-layer game model is solved by the Nash equilibrium method.

[0014] The inputs to the weighted thermal comfort function include the predicted average thermal sensation index, population density index, air supply angle of the electric side-supply nozzle, air supply velocity of the electric side-supply nozzle, and air volume of the bottom air supply nozzle for each monitoring area. The output is a weighted thermal comfort evaluation index, which is calculated by multiplying the absolute value of the deviation of the predicted average thermal sensation index of each monitoring area by the population density weight of the corresponding monitoring area, summing the results, and taking the negative value.

[0015] The inputs to the air supply energy consumption function include the power of the air supply fan, the total air volume, and the static pressure loss. The output is the normalized air supply energy consumption, which is calculated by multiplying the power of the air supply fan by the operating time. The power of the air supply fan is proportional to the product of the total air volume and the static pressure loss.

[0016] Among them, the heat exchange optimization algorithm for minimizing thermodynamic entropy production is based on the second law of thermodynamics to establish an irreversible loss model of the heat exchanger. The irreversible loss includes two parts: heat transfer temperature difference entropy production and flow resistance entropy production. A heat exchange optimization model is established with the goal of minimizing the total entropy production. The decision variables are the refrigerant flow rate of the heat exchanger and the target outlet air temperature of the heat exchanger. The constrained optimization problem is transformed into an unconstrained problem using the Lagrange multiplier method, and the problem is solved iteratively by the gradient descent algorithm.

[0017] In the airflow field reconstruction model, the aggregation weight coefficient of the graph convolution layer is determined based on the personnel density index, the air volume supplied by the lower air outlet, and the distance between spatial nodes. The formula for calculating the aggregation weight coefficient is the product of the personnel density index divided by the standard personnel density and the air volume supplied by the lower air outlet divided by the standard air volume, and then multiplied by the negative exponential function of the distance between spatial nodes divided by the standard distance.

[0018] The comfort deviation threshold is dynamically set according to the activity type and clothing of people in the room. For office workers who are sitting and wearing standard office clothing, the comfort deviation threshold is set to 0.3. For people in commercial places who are walking and wearing light clothing, the comfort deviation threshold is set to 0.5. For people in industrial places who are engaged in heavy physical labor, the comfort deviation threshold is set to 0.7.

[0019] The feedback correction mechanism is implemented when the absolute value of the actual predicted average thermal sensation index deviation is greater than the comfort deviation threshold. The predicted average thermal sensation index deviation is used as a feedback correction parameter to adjust the objective function of the upper-level game model of the air supply regulation two-level game model, increasing the weight coefficient of the predicted average thermal sensation index deviation. This makes the air supply regulation two-level game model pay more attention to the comfort target in the next solution. Through multiple iterations, the predicted average thermal sensation index deviation is gradually reduced to form a closed-loop control system.

[0020] This invention employs an intelligent adjustment method combining a graph neural network-based airflow field reconstruction model with a two-layer game theory model for air supply regulation. This solves the technical problem of distributed air handling equipment struggling to adjust air supply parameters in real-time based on dynamic changes in personnel distribution to simultaneously ensure thermal comfort and energy efficiency. The airflow field reconstruction model discretizes a three-dimensional space into several spatial nodes and constructs a spatial topology graph. It utilizes graph convolutional layers to aggregate airflow transmission information between adjacent nodes and combines sine and cosine position coding to enable the network to automatically perceive spatial positional relationships, achieving millisecond-level rapid prediction of temperature and velocity field distributions. This overcomes the time-consuming nature of traditional numerical simulation methods for solving the Navier-Stokes equations. The two-layer game theory model for air supply regulation is based on the predicted three-dimensional spatial flow field distribution data. The upper-layer game theory model maximizes the weighted thermal comfort in densely populated areas, while the lower-layer model minimizes air supply energy consumption. The Nash equilibrium method is used to obtain the optimal combination of air supply parameters that balances comfort and energy efficiency. A feedback correction mechanism dynamically adjusts the weight coefficients of the game theory model based on the actual deviation of the predicted average thermal sensation index, forming a closed-loop control system that gradually reduces comfort deviation. In summary, the present invention solves the technical problem mentioned in the background art that distributed air handling equipment is unable to adjust the air supply parameters in real time according to the dynamic changes in personnel distribution to simultaneously ensure thermal comfort and energy efficiency. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention.

[0022] Figure 2 This is a schematic diagram of a distributed air handling unit.

[0023] Figure 3 Encode the feature distribution map of spatial node locations.

[0024] Figure 4 This is a distribution map of population density index in each monitoring area.

[0025] Figure 5 This is a graph showing the predicted change in the average thermal sensation index during air supply adjustment.

[0026] In the attached drawings, the reference numerals are explained as follows: 1. Return air grille; 2. Return air filter; 3. Thermal comfort sensor; 4. Return water pipe; 5. Inlet water pipe; 6. Soundproof casing; 7. Supply air silencer; 8. Electric side supply nozzle; 9. Downward air supply outlet; 10. Insulated wall panel; 11. Supply fan; 12. Heat exchanger; 13. Water collection tray; 14. Return air box; 15. Fresh air main duct; 16. Displacement air supply outlet; 17. Inspection door; 18. Intelligent control cabinet; 19. Air diversion device; 20. Fire extinguisher box; 21. Personnel detection camera. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below.

[0028] like Figure 1 The diagram shown is a flowchart of an intelligent adjustment method for a distributed air handling device provided by the present invention. This method includes the following steps:

[0029] S01. After the distributed air handling equipment is started, the distribution images of people in the area around the equipment are collected by the personnel detection camera, and the dry bulb temperature, relative humidity, average radiation temperature and air velocity data of the personnel activity area are collected by the thermal comfort sensor.

[0030] S02. Input the personnel distribution image into the airflow field reconstruction model. The airflow field reconstruction model outputs three-dimensional spatial flow field distribution data, which includes the temperature field distribution and velocity field distribution of each node around the equipment.

[0031] S03. Calculate the population density index of each monitoring area based on the population distribution image, and calculate the predicted average thermal sensation index of the population activity area based on the dry bulb temperature, relative humidity, mean radiation temperature and air velocity data.

[0032] S04. Based on the personnel density index, the predicted average thermal sensation index value and the three-dimensional spatial flow field distribution data, a two-layer game model for air supply regulation is established to solve for the air supply angle of the electric side air supply nozzle, the air supply velocity of the electric side air supply nozzle and the air supply volume of the lower air supply nozzle.

[0033] S05. Based on the air delivery angle of the electric side nozzle, the air delivery speed of the electric side nozzle, and the air delivery volume of the lower air outlet, control the electric side nozzle actuator to adjust the air delivery angle and air delivery speed, and control the lower air outlet damper to adjust the air delivery volume.

[0034] S06. Based on the temperature field distribution, combined with the outdoor fresh air temperature and return air temperature, calculate the target outlet air temperature and refrigerant flow rate of the heat exchanger using the heat exchange optimization algorithm that minimizes thermodynamic entropy production, and control the water flow regulating valve of the heat exchanger to adjust the flow rate.

[0035] S07. Within 15 to 30 seconds after step S05 is executed, re-acquire images of personnel distribution, as well as data on dry bulb temperature, relative humidity, mean radiant temperature and air velocity, and calculate the actual predicted mean thermal sensation index value.

[0036] S08. Calculate the predicted average thermal sensation index deviation between the actual predicted average thermal sensation index value and the target predicted average thermal sensation index value. When the absolute value of the predicted average thermal sensation index deviation value is greater than the comfort deviation threshold, return the predicted average thermal sensation index deviation value as a feedback correction parameter to step S04. When the absolute value of the predicted average thermal sensation index deviation value is less than or equal to the comfort deviation threshold, return to step S01.

[0037] Four personnel detection cameras are installed in four directions around the distributed air handling unit to collect information on the distribution of people within a 15m radius of the unit. Thermal comfort sensors are installed on the indoor seats to collect environmental parameters of the areas where people are active.

[0038] The airflow field reconstruction model is structured as a spatial flow field prediction network based on a graph neural network. The three-dimensional space covered by the distributed air handling equipment is discretized into several spatial nodes. The airflow transmission relationship between adjacent spatial nodes is represented as directed edges, constructing a spatial topological graph structure. The coordinates of the spatial nodes are embedded into a high-dimensional feature space through sine and cosine position encoding, maintaining the relative spatial positions of the nodes. The spatial flow field prediction network contains three graph convolutional layers and two fully connected layers. The first graph convolutional layer has an input dimension of 128 and an output dimension of 256; the second graph convolutional layer has an input dimension of 256 and an output dimension of 512; the third graph convolutional layer... The layered input dimension is 512 and the output dimension is 256. Two fully connected layers map the 256-dimensional features to 128-dimensional features and the final output dimension, respectively. The output dimension includes four channels: temperature component and velocity components in three directions. Residual connections are used to add the input features to the output of each layer, accelerating the training and convergence of the spatial flow field prediction network. The sine and cosine position encoding maps the three-dimensional coordinates x, y, and z of the spatial nodes to high-dimensional vectors through sine and cosine functions of different frequencies. This allows the spatial flow field prediction network to automatically perceive the spatial positional relationship of the spatial nodes when learning the airflow transmission law, avoiding feature confusion caused by spatial translation.

[0039] The steps for establishing the training dataset for the airflow field reconstruction model specifically include: building a distributed air handling equipment model with a scale of 1:5 in a laboratory environment; arranging 200 temperature sensors and 200 wind speed sensors around the distributed air handling equipment model to form a dense grid of measuring points; collecting time series data of temperature and wind speed at each measuring point under different air supply conditions; continuously collecting data for 600 seconds under each air supply condition at a sampling frequency of 1Hz; changing the air supply angle, air supply speed, personnel distribution pattern, and obstacle position; collecting a total of 500 sets of flow field data under different air supply conditions; randomly selecting 20% ​​of the measuring points in each set of flow field data as known node inputs; and using the remaining 80% of the measuring points as prediction target outputs to form training sample pairs; and dividing the 500 sets of flow field data under different air supply conditions into a training set, a validation set, and a test set in a ratio of 8:1:1.

[0040] The specific steps for training the airflow field reconstruction model include: using mean squared error as the loss function to measure the difference between the predicted flow field and the actual flow field; using the Adam optimizer to update the parameters; setting the initial learning rate to 0.001; decreasing the learning rate to 0.5 times every 30 training cycles; setting the batch size to 32; and setting the total number of training cycles to 200. During training, the loss function value on the validation set is monitored, and training is stopped early when the loss function value on the validation set no longer decreases after 20 consecutive training cycles. The model parameters with the smallest loss function value on the validation set are saved as the final model parameters.

[0041] The implementation of the position-encoded sequence position-aware framework in the airflow field reconstruction model is as follows: the three-dimensional coordinates x, y, and z of the spatial nodes are encoded using sine and cosine functions with different frequency parameters, respectively. For the x-coordinate component, frequencies of 1, 10, and 100 are used. The system uses six encoding channels for the sine and cosine functions. The y-coordinate and z-coordinate are encoded in the same way. The position encodings of the three coordinate components are concatenated to form an 18-dimensional position feature vector. This 18-dimensional position feature vector is concatenated with the physical features of the spatial nodes, including the initial temperature and initial wind speed, and then used as the input to the graph convolutional layer. In the graph convolution operation, the spatial node features are aggregated by weighting the information of adjacent spatial nodes through the adjacency matrix. The position encoding enables the spatial flow field prediction network to distinguish the influence weight of spatial nodes at different spatial locations on the current spatial node during the aggregation process. Larger aggregation weights are assigned to spatial nodes that are closer, and smaller aggregation weights are assigned to spatial nodes that are farther away. This spatial distance perception capability makes the airflow diffusion law learned by the spatial flow field prediction network conform to physical intuition.

[0042] The position-encoded sequence position-aware framework brings three main technical benefits to the airflow field reconstruction model: First, in terms of spatial generalization ability, position encoding enables the airflow field reconstruction model to handle spatial configurations not seen during the training phase. When the installation location of distributed air handling equipment or the room layout changes in practical applications, the airflow field reconstruction model can adapt to the new spatial topology without retraining, because position encoding maps absolute coordinates to relative positional relationships, maintaining spatial translation invariance. Second, in terms of prediction accuracy, position encoding helps the spatial flow field prediction network capture the anisotropic characteristics of airflow propagation. In the vertical direction, due to the buoyancy effect, the upward flow speed of hot air... The gradient is significantly different from the horizontal direction. Position encoding, through independent encoding of different coordinate components, enables the spatial flow field prediction network to learn this direction-dependent propagation law, significantly improving the prediction accuracy of temperature and velocity field distributions. Finally, in terms of computational efficiency, position encoding avoids the iterative process of solving the Navier-Stokes equations in traditional numerical simulation methods. The forward propagation of the airflow field reconstruction model only requires matrix multiplication and activation function operations. The time taken for a single flow field reconstruction is reduced from several minutes in traditional computational fluid dynamics methods to milliseconds, meeting the response speed requirements of real-time control. This allows distributed air handling equipment to dynamically adjust its air supply strategy according to rapid changes in personnel distribution.

[0043] The method for calculating the population density index is as follows: The population distribution image captured by the personnel detection camera is divided into several monitoring sub-regions. Each monitoring sub-region corresponds to a fan-shaped area around the device, with a central angle of 90 degrees and a radius of 15 meters. Human detection is performed on the image of each monitoring sub-region to identify the number of people within that sub-region. The number of people is then divided by the area of ​​the monitoring sub-region to obtain the population density index. The unit of the population density index is people / ... .

[0044] The predicted average thermal perception index is the PMV value calculated according to the international standard ISO 7730. The calculation formula comprehensively considers six parameters: human metabolic rate, clothing thermal resistance, dry-bulb temperature, mean radiant temperature, air velocity, and relative humidity. It is obtained by solving the human body's thermal balance equation to obtain a dimensionless value characterizing human temperature sensation. A predicted average thermal perception index of 0 indicates neutral thermal sensation, a negative value indicates slightly cold, and a positive value indicates slightly hot. The target predicted average thermal perception index is set to 0.

[0045] The two-layer game model for air supply regulation includes an upper-layer game model that aims to maximize comfort and a lower-layer game model that aims to minimize energy consumption.

[0046] The objective function of the upper-level game model is to maximize the weighted thermal comfort function. The inputs to the weighted thermal comfort function include the predicted average thermal sensation index for each monitoring area, the population density index, the air delivery angle of the electrically powered side-mounted nozzles, the air velocity of the electrically powered side-mounted nozzles, and the air volume of the downward-mounted nozzles. The output is a weighted thermal comfort evaluation index, which is calculated by multiplying the absolute value of the deviation of the predicted average thermal sensation index for each monitoring area by the corresponding population density weight, summing the results, and taking the negative value. Monitoring areas with higher population density are assigned a higher weight. The objective function of the upper-level game model, with a high population density weight, is expressed as follows: The objective function value of the upper-level game model equals the sum of the products of the negative predicted average thermal comfort index deviation and the population density weight for each monitoring area, divided by the product of the standard thermal comfort deviation and the standard population density weight. Here, the predicted average thermal comfort index deviation is the absolute value of the difference between the predicted average thermal comfort index value and the target predicted average thermal comfort index value; the population density weight is the population density index divided by the standard population density; the standard thermal comfort deviation is taken as 0.5; and the standard population density is taken as 0.1 people / ... The standard personnel density weight is set to 1; the constraints of the upper-level game model include an air delivery angle range of -45 degrees to +45 degrees for the electric side-supply nozzles, an air velocity range of 2 m / s to 8 m / s for the electric side-supply nozzles, and an air volume range of 2000 for the lower air supply nozzles. / h to 10000 / h.

[0047] The objective function of the lower-level game model is to minimize the air supply energy consumption function. The inputs to this function include the blower power, total air volume, and static pressure loss. The output is the normalized air supply energy consumption, which is calculated by multiplying the blower power by the operating time. The blower power is proportional to the product of the total air volume and the static pressure loss. The objective function of the lower-level game model is expressed as follows: the objective function value equals the product of the total air volume and the static pressure loss divided by the product of the standard total air volume and the standard static pressure loss. The total air volume is the sum of the air volume at the electric side nozzle corresponding to the air velocity and the air volume at the lower air outlet. The static pressure loss is determined by the total air volume based on the duct resistance characteristic curve. The standard total air volume is set to 10000. / h, the standard static pressure loss is taken as 250Pa; the constraints of the lower-level game model include that the total air supply volume must meet the requirement that the indoor air exchange rate is not less than 6 times / h and the air velocity in each monitoring area does not exceed 0.5m / s.

[0048] The two objective functions of the two-layer game model for air supply regulation are related through a coupling term. The coupling term is the integral of the product of the temperature gradient and velocity gradient of each spatial node in the three-dimensional spatial flow field distribution data. The coupling term simultaneously affects the weighted thermal comfort evaluation index of the upper-layer game model and the normalized air supply energy consumption of the lower-layer game model. Larger temperature and velocity gradients mean stronger airflow mixing, which is conducive to quickly reaching the target temperature and improving comfort, but at the same time, it leads to greater turbulent dissipation and increased energy consumption. The two-layer game model for air supply regulation is solved by the Nash equilibrium method to find the air supply parameter combination that makes the objective functions of the upper-layer game model and the lower-layer game model reach a relatively optimal state simultaneously.

[0049] The heat transfer optimization algorithm that minimizes thermodynamic entropy production establishes an irreversible loss model for the heat exchanger based on the second law of thermodynamics. Irreversible losses include two parts: heat transfer temperature difference entropy production and flow resistance entropy production. The heat transfer temperature difference entropy production is caused by the temperature difference between the refrigerant side and the air side. A larger temperature difference results in a faster heat transfer rate but also a larger heat transfer temperature difference entropy production. The rate of heat transfer temperature difference entropy production equals the heat transfer multiplied by the difference between the reciprocal of the reciprocal of the refrigerant side temperature and the reciprocal of the air side temperature, divided by the standard temperature. The flow resistance entropy production is caused by the pressure loss of the fluid flow. The pressure loss is proportional to the square of the flow velocity. The rate of flow resistance entropy production equals the volumetric flow rate multiplied by the pressure loss, divided by the fluid temperature, and then divided by the standard pressure.

[0050] A heat exchange optimization model is established with the goal of minimizing total entropy production. The decision variables are the refrigerant flow rate of the heat exchanger and the target outlet air temperature of the heat exchanger. The constraints include that the heat exchange capacity must meet the indoor heat load demand, the refrigerant flow rate of the heat exchanger is within the range of 50% to 100% of the rated flow rate, and the target outlet air temperature of the heat exchanger is not lower than the dew point temperature. The constrained optimization problem is transformed into an unconstrained problem using the Lagrange multiplier method. Lagrange multipliers are introduced to attach constraints to the objective function of the heat exchanger optimization model. Partial derivatives are calculated with respect to the decision variables and the Lagrange multipliers, and then set to zero to obtain the Kuhn-Tark condition equations. Since the physical properties of the refrigerant and air change nonlinearly with temperature, piecewise linearization is applied to these parameters, dividing the temperature range into several temperature sub-intervals. Within each sub-interval, the physical properties are approximately linear functions. The Kuhn-Tark condition equations are iteratively solved using a gradient descent algorithm to obtain the optimal refrigerant flow rate and target outlet air temperature of the heat exchanger. The step size of the gradient descent algorithm is adaptively adjusted based on the eigenvalues ​​of the Hessian matrix of the objective function of the heat exchanger optimization model.

[0051] The aggregation weight coefficient of the graph convolutional layer in the airflow field reconstruction model is determined based on the population density index, the air volume supplied from the lower air outlet, and the distance between spatial nodes. The formula for calculating the aggregation weight coefficient is as follows: The aggregation weight coefficient equals the product of the population density index divided by the standard population density and the air volume supplied from the lower air outlet divided by the standard air volume, multiplied by the negative exponential function of the distance between spatial nodes divided by the standard distance, where the standard population density is taken as 0.1 people / ... The standard air supply volume is 10000. / h, with a standard distance of 1m and an exponent of -2 for the negative exponential function, the aggregation weight coefficient enables the airflow field reconstruction model to give spatial nodes a stronger feature aggregation capability in monitoring areas with dense populations and large airflow from the down-ventilation outlet, thereby enhancing the prediction accuracy of airflow diffusion.

[0052] The comfort deviation threshold is dynamically set according to the activity type and clothing of the people in the room. For office workers who are sitting and wearing standard office clothing, the comfort deviation threshold is set to 0.3. For people in commercial places who are walking and wearing lighter clothing, the comfort deviation threshold is set to 0.5. For people in industrial places who are engaged in heavy physical labor, the comfort deviation threshold is set to 0.7.

[0053] The replacement air inlet is located at the bottom of the distributed air handling unit. Outdoor fresh air is delivered into the room at a low speed of 0.2 m / s to 0.4 m / s via a guide vane with an angle of 15 to 25 degrees. The outdoor fresh air mixes with the circulating air from the downflow vent near the ground to form a uniformly heated airflow layer. The outdoor fresh air volume is determined based on the indoor... Dynamic concentration adjustment, when indoor When the concentration exceeds 1000ppm, increase the outdoor fresh air volume; when the indoor... Reduce the outdoor fresh air volume when the concentration is below 800ppm.

[0054] The intelligent control cabinet of the distributed air handling unit receives data from personnel detection cameras and thermal comfort sensors, runs an airflow field reconstruction model and a two-layer game model for air supply regulation, and outputs control commands to the electric side-supply nozzle actuator, the bottom air supply valve, and the heat exchanger water flow regulating valve. The intelligent control cabinet is also linked with the building fire protection system. When the indoor temperature exceeds 60°C, it is determined that a fire has occurred, and the return air duct is immediately closed, the smoke exhaust mode is activated, and an alarm message is sent to the building management system.

[0055] The feedback correction mechanism is implemented in step S08. When the absolute value of the predicted average thermal sensation index deviation is greater than the comfort deviation threshold, the predicted average thermal sensation index deviation is used as a feedback correction parameter to adjust the objective function of the upper-level game model of the air supply regulation two-layer game model. The weight coefficient of the predicted average thermal sensation index deviation is increased, so that the air supply regulation two-layer game model pays more attention to the comfort target in the next solution. Through multiple iterations, the predicted average thermal sensation index deviation is gradually reduced, forming a closed-loop control system.

[0056] The specific implementation methods of the above steps are described in detail below.

[0057] The specific implementation of step S01 involves using four personnel detection cameras to collect real-time images of personnel distribution within a 15m radius around the distributed air handling equipment. The cameras employ visible light imaging sensors with a resolution of 1920×1080 pixels and a frame rate of 30 frames per second to ensure continuous dynamic information capture during personnel movement. Simultaneously, thermal comfort sensors installed on indoor seats collect data on dry-bulb temperature, relative humidity, mean radiant temperature, and air velocity in the personnel activity area. The thermal comfort sensors have a sampling frequency of 1Hz, a dry-bulb temperature measurement range of 10℃ to 40℃ with a measurement accuracy of ±0.1℃, a relative humidity measurement range of 0% to 100% with a measurement accuracy of ±2%, a mean radiant temperature measured using a black sphere thermometer, and an air velocity measurement range of 0m / s to 2m / s with a measurement accuracy of ±0.01m / s. This data acquisition process provides raw input parameters for subsequent personnel density calculations and thermal comfort assessments.

[0058] The specific implementation of step S02 involves inputting the personnel distribution image into the airflow field reconstruction model. First, the personnel distribution image is preprocessed to extract the personnel position coordinates as known heat source distribution information. Then, the three-dimensional space covered by the distributed air handling equipment is discretized into a regular grid with a grid node spacing of 0.5m. Each spatial node carries initial temperature and initial wind speed characteristics. The three-dimensional coordinates x, y, and z of the spatial nodes are mapped into high-dimensional feature vectors using sine and cosine position encoding. For the x-coordinate component, frequencies of 1, 10, and 100 are used. The system has six encoding channels for the sine and cosine functions. The y-coordinate and z-coordinate are encoded in the same way. The position encodings of the three coordinate components are concatenated to form an 18-dimensional position feature vector. The 18-dimensional position feature vector is concatenated with the physical features of the spatial nodes and then input into a graph convolutional layer. The graph convolutional layer performs weighted aggregation of the features of adjacent spatial nodes through an adjacency matrix. The aggregation weight coefficients are dynamically calculated based on the distance between spatial nodes and the personnel density index. After forward propagation through three layers of graph convolutional layers and two layers of fully connected layers, the system outputs three-dimensional spatial flow field distribution data, including the temperature field distribution and velocity field distribution of each spatial node around the equipment. The airflow field reconstruction model adopts a spatial flow field prediction network based on a graph neural network. It uses a graph structure to express the airflow transmission relationship between spatial nodes and maintains the relative relationship of spatial positions through position encoding, so that the model can adapt to different spatial configurations without retraining.

[0059] The specific implementation of step S03 involves calculating the population density index of each monitoring area based on the population distribution image. The population distribution image captured by the personnel detection camera is divided into four monitoring sub-areas. Each monitoring sub-area corresponds to a fan-shaped area around the device, with a central angle of 90 degrees and a radius of 15m. A human detection algorithm based on a deep convolutional neural network is applied to the image of each monitoring sub-area to identify the number of people within the monitoring sub-area. The population density index is obtained by dividing the number of people by the area of ​​the monitoring sub-area, with units of people / Then, based on the dry-bulb temperature, relative humidity, mean radiant temperature, and air velocity data collected by the thermal comfort sensor, combined with the human metabolic rate and clothing thermal resistance parameters, the predicted average thermal sensation index is calculated using the human body thermal balance equation specified in the international standard ISO7730. The human body thermal balance equation comprehensively considers the dynamic balance between human heat production and heat dissipation. The predicted average thermal sensation index is obtained by iteratively solving a set of nonlinear equations. The iteration termination condition is set to the difference between two adjacent iteration results being less than 0.01. The population density index reflects the degree of population gathering in different areas, and the predicted average thermal sensation index reflects the subjective thermal sensation state of people. Both serve as the basis for air supply adjustment.

[0060] The specific implementation of step S04 involves establishing a two-layer game model for air supply regulation based on the population density index, the predicted average thermal sensation index, and the three-dimensional spatial flow field distribution data. This two-layer game model includes an upper-layer game model and a lower-layer game model. The upper-layer game model aims to maximize comfort, with the objective function being a weighted thermal comfort function. It involves multiplying the absolute value of the predicted average thermal sensation index deviation for each monitoring area by the corresponding population density weight, summing the results, and taking the negative value. The population density weight is calculated by dividing the population density index by the standard population density, where the standard population density is 0.1 people / ... The lower-level game model aims to minimize energy consumption. The objective function is the normalized supply air energy consumption, calculated by dividing the product of the total supply air volume and static pressure loss by the product of the standard total supply air volume and standard static pressure loss. The two objective functions are linked through a coupling term, which is the spatial integral of the product of the temperature gradient and velocity gradient at each spatial node in the three-dimensional spatial flow field distribution data. The Nash equilibrium method is used to solve the two-level game model of supply air regulation. The Nash equilibrium method seeks to find a solution where unilateral changes in strategy by either player result in... For strategy combinations that cannot achieve better results, an iterative optimization algorithm is used to solve the problem. The upper-level game model fixes the decision variables of the lower-level game model to solve for the optimal air supply angle and air supply velocity. The lower-level game model fixes the decision variables of the upper-level game model to solve for the optimal air supply volume. The process is repeated until the decision variables of the two models converge. The convergence criterion is set to the change in decision variables between two adjacent iterations being less than 1%. Finally, the air supply angle, air supply velocity, and air supply volume of the lower air supply outlet are obtained.

[0061] The specific implementation of step S06 involves calculating the target outlet air temperature and refrigerant flow rate of the heat exchanger using a heat exchange optimization algorithm that minimizes thermodynamic entropy production, based on the temperature field distribution and the outdoor fresh air and return air temperatures. This algorithm establishes an irreversible loss model for the heat exchanger based on the second law of thermodynamics. Irreversible losses include heat transfer temperature difference entropy production and flow resistance entropy production. The heat transfer temperature difference entropy production is caused by the temperature difference between the refrigerant side and the air side, while the flow resistance entropy production is caused by the pressure loss of the fluid flow. A heat exchange optimization model is established with the goal of minimizing total entropy production. The decision variables are the refrigerant flow rate and the target outlet air temperature of the heat exchanger. The constrained optimization problem is transformed into an unconstrained problem using the Lagrange multiplier method. The Lagrange multiplier is introduced to ensure that the heat exchange meets the indoor heat load demand, and that the refrigerant flow rate of the heat exchanger is within the range of 50% to 100% of the rated flow rate. Constraints such as the target outlet air temperature of the heat exchanger not being lower than the dew point temperature are added to the objective function of the heat exchanger optimization model. Partial derivatives are taken with respect to the decision variables and the Lagrange multiplier and set to zero to obtain the Kuhn-Tak condition equations. Since the physical properties of the refrigerant and air change nonlinearly with temperature, piecewise linearization is applied to the physical properties, dividing the temperature range into 10 temperature sub-intervals. Within each temperature sub-interval, the physical properties are approximately linear functions. The Kuhn-Tak condition equations are solved iteratively using the gradient descent algorithm. The step size of the gradient descent algorithm is adaptively adjusted according to the eigenvalues ​​of the Hessian matrix of the objective function of the heat exchanger optimization model. The initial step size is set to 0.1, and the step size is dynamically adjusted according to the descent rate of the objective function after each iteration. The iteration termination condition is set to the change in the objective function being less than 0.001 or the number of iterations exceeding 100. Finally, the optimal refrigerant flow rate and the target outlet air temperature of the heat exchanger are obtained.

[0062] The specific implementation of step S08 involves calculating the predicted average thermal sensation index deviation between the actual predicted average thermal sensation index value and the target predicted average thermal sensation index value. The target predicted average thermal sensation index value is set to 0, and the predicted average thermal sensation index deviation is the difference between the actual predicted average thermal sensation index value and the target predicted average thermal sensation index value. It is then determined whether the absolute value of the predicted average thermal sensation index deviation is greater than a comfort deviation threshold. The comfort deviation threshold is dynamically set based on the activity type and clothing of the people indoors. The comfort deviation threshold is 0.3 for office spaces, 0.5 for commercial spaces, and 0.7 for industrial spaces. When the predicted average thermal sensation index deviation value... When the absolute value is greater than the comfort deviation threshold, the predicted average thermal sensation index deviation is used as a feedback correction parameter and returned to step S04. The objective function of the upper-level game model of the air supply regulation two-layer game model is adjusted, and the weight coefficient of the predicted average thermal sensation index deviation is increased. The weight coefficient increment is set to 0.1, so that the air supply regulation two-layer game model pays more attention to the comfort target in the next solution. Through multiple iterations, the predicted average thermal sensation index deviation is gradually reduced to form a closed-loop control system. When the absolute value of the predicted average thermal sensation index deviation is less than or equal to the comfort deviation threshold, it indicates that the current air supply strategy has met the comfort requirements, and the process returns to step S01 to continue monitoring changes in personnel distribution and environmental parameters.

[0063] It should be noted that the key technical ideas of this invention include an airflow field reconstruction model based on graph neural networks, an air supply regulation strategy based on a two-layer game model, and a heat transfer optimization algorithm that minimizes thermodynamic entropy production. The airflow field reconstruction model based on graph neural networks discretizes the three-dimensional space into a node grid, uses a graph structure to express the airflow transmission relationship between adjacent nodes, and combines sine and cosine position encoding to maintain the relative spatial positions of the nodes. This allows the model to predict the complete three-dimensional flow field distribution from sparsely distributed sensor measurement data. Compared to the complex iterative process of solving the Navier-Stokes equations required by traditional computational fluid dynamics methods, the airflow field reconstruction model reduces the time required for a single flow field reconstruction from several minutes to milliseconds, meeting the response speed requirements of real-time control. Simultaneously, position encoding endows the model with spatial generalization ability; when the installation location of distributed air handling equipment or the room layout changes in practical applications, the model can adapt to the new spatial topology without retraining. The two-layer game theory model for air supply regulation aims to maximize comfort in the upper layer and minimize energy consumption in the lower layer. These two objective functions are linked by a coupling term, which is the spatial integral of the product of the temperature gradient and velocity gradient, reflecting the combined impact of airflow mixing intensity on comfort and energy consumption. The Nash equilibrium method is used to solve the two-layer game theory model, seeking the optimal combination of air supply parameters that simultaneously achieves relative best comfort and energy consumption. Compared to traditional methods that optimize only one objective, leading to a severe deterioration of the other, the two-layer game theory model achieves a dynamic balance between comfort and energy consumption. The heat exchange optimization algorithm for minimizing thermodynamic entropy production establishes an irreversible loss model for the heat exchanger based on the second law of thermodynamics, including entropy production from heat transfer temperature difference and flow resistance. It aims to minimize total entropy production to find the optimal refrigerant flow rate and outlet air temperature, ensuring maximum thermodynamic perfection of the heat exchange process. Compared to traditional methods that control heat exchanger operation solely based on temperature difference, the entropy production minimization method optimizes the heat exchange process from an energy quality perspective, reducing irreversible losses and improving system energy efficiency. The synergistic effect of the three key technical approaches is reflected in the fact that the airflow field reconstruction model provides accurate three-dimensional flow field distribution data for the two-layer game model, enabling the game model to predict the impact of different air supply strategies on indoor airflow organization. The air supply parameters obtained from the two-layer game model are further used as the basis for calculating the aggregation weight coefficients of the graph convolution layer in the airflow field reconstruction model, forming a positive feedback optimization mechanism. At the same time, the heat exchange optimization algorithm adjusts the heat exchanger operating parameters according to the temperature gradient in the flow field distribution to ensure dynamic matching between the air supply temperature and the indoor heat load. The three work together to construct a complete control closed loop from airflow field prediction to air supply regulation to heat exchange optimization. Compared with the traditional method where each subsystem operates independently and lacks coordination, this invention achieves efficient and energy-saving operation of distributed air handling equipment under dynamic changes in personnel distribution through multi-level and multi-objective collaborative optimization.

[0064] It should be noted that this invention also solves the following technical problem: traditional airflow field prediction methods struggle to handle changes in spatial configuration, leading to insufficient generalization ability. When the installation location of distributed air handling equipment or the room layout changes, traditional numerical simulation-based methods require rebuilding the computational domain grid and adjusting boundary conditions, which is time-consuming and complex. Data-driven neural network methods, due to training data only covering specific spatial configurations, experience a significant decrease in prediction accuracy when facing new spatial topologies. This invention introduces sine and cosine position encoding into the airflow field reconstruction model, mapping the absolute coordinates of spatial nodes to relative positional relationships, maintaining spatial translation invariance. This allows the model to handle spatial configurations not seen during the training phase. When the spatial layout changes in practical applications, the model can adapt to new spatial topologies without retraining, significantly improving the spatial generalization ability of the airflow field reconstruction model and solving the technical problem of traditional methods' inability to cope with changes in spatial configuration.

[0065] Furthermore, this invention addresses the technical problem of traditional heat exchanger control methods struggling to balance heat exchange efficiency and flow resistance losses. Traditional heat exchanger control methods typically focus only on whether the outlet air temperature reaches the set value, adjusting the refrigerant flow rate through simple temperature feedback. This ignores the impact of heat transfer temperature difference entropy production and flow resistance entropy production on irreversible system losses during heat exchange. When the refrigerant flow rate is too high, although the heat exchange rate is fast, the flow resistance loss increases significantly. When the refrigerant flow rate is too low, the increased heat transfer temperature difference leads to a decrease in heat transfer efficiency. This invention employs a heat exchanger optimization algorithm that minimizes thermodynamic entropy production. Based on the second law of thermodynamics, it establishes an irreversible loss model that includes heat transfer temperature difference entropy production and flow resistance entropy production. It optimizes the refrigerant flow rate and target outlet air temperature with the goal of minimizing total entropy production. The constrained optimization problem is transformed into an unconstrained problem using the Lagrange multiplier method, and the gradient descent algorithm is used for iterative solution. This achieves synergistic optimization of heat exchange efficiency and flow resistance losses while meeting indoor heat load requirements, solving the technical problem of traditional methods struggling to balance heat exchange efficiency and flow resistance losses.

[0066] Specifically, the principle of this invention is as follows: The technical solution of this invention can solve the above-mentioned core technical problems by using personnel distribution images collected by personnel detection cameras as input to the airflow field reconstruction model. The airflow field reconstruction model represents spatial nodes and their connections as a graph structure based on a graph neural network architecture. The graph convolutional layer aggregates the features of adjacent nodes through adjacency matrix weighting. Sine and cosine position encoding maps the three-dimensional coordinates of spatial nodes into high-dimensional feature vectors and concatenates them with physical features into the network. This allows the network to distinguish the influence weights of nodes at different spatial locations when learning the airflow transmission law, assigning larger aggregation weights to nodes that are closer and smaller aggregation weights to nodes that are farther away. This spatial distance perception capability makes the airflow diffusion law learned by the model conform to physical intuition. The forward propagation of the model only requires matrix multiplication and activation function operations to output three-dimensional spatial flow field distribution data, reducing the computation time from several minutes in traditional methods to... With millisecond-level response times, meeting real-time control requirements, the two-layer game model for air supply regulation combines predicted flow field distribution data with population density index and predicted average thermal comfort index. The upper-layer game model prioritizes comfort needs in densely populated areas by maximizing the weighted thermal comfort function, while the lower-layer game model reduces energy consumption while meeting comfort constraints by minimizing the air supply energy consumption function. The two-layer game models are linked by the spatial integral of the product of temperature gradient and velocity gradient as a coupling term. The Nash equilibrium solution method seeks the air supply parameter combination that makes both objective functions reach a relatively optimal state simultaneously. The feedback correction mechanism adjusts the weight coefficients of the upper-layer game model based on the actual deviation of the predicted average thermal comfort index, and iterates multiple times to gradually reduce comfort deviation, forming a closed-loop control. Therefore, the technical solution of this invention can logically predict the three-dimensional spatial flow field distribution and optimize air supply parameters in real time according to the dynamic changes in population distribution, while ensuring thermal comfort and energy efficiency.

[0067] The following provides a specific embodiment 1 of the present invention. The specific implementation methods of steps S01, S02, S05 and S07 in this embodiment 1 are the same as those described above, and will not be repeated in detail here. The specific implementation methods of other steps are described in detail below.

[0068] The specific implementation of step S03 is as follows: the formula for calculating the population density index is expressed as follows:

[0069] ;

[0070] In the formula, For the first Population density index of each monitored sub-area, in person / ; For the first The number of people identified in each monitoring sub-region, in units of individuals; For the first The area of ​​each monitoring sub-region, in units of ; The monitoring sub-regions are numbered 1, 2, 3, and 4. The predicted average thermal sensation index value is... Calculated using the international standard ISO 7730, taking into account the human metabolic rate. thermal resistance of clothing Dry bulb temperature Mean radiant temperature air velocity and relative humidity Six parameters, including the human body's thermal balance equation, are used to obtain dimensionless values ​​characterizing human body's hot and cold sensations. The target is to predict the average thermal sensation index. Set to 0. Wherein, Human metabolic rate, unit: W / According to the metabolic rate table, the office status experience value is 70. This refers to the thermal resistance of clothing, measured in Clo (clo). It can be obtained from a clothing thermal resistance table, with a standard office uniform value of 1.0. Dry bulb temperature, in °C, is directly collected by the thermal comfort sensor; The average radiant temperature, in °C, is directly collected by the thermal comfort sensor. The air velocity is measured in m / s and is directly collected by the thermal comfort sensor. The relative humidity is expressed as a percentage (%) and is directly collected by the thermal comfort sensor.

[0071] The specific implementation of step S04 is as follows: the objective function of the upper-level game model is expressed as follows:

[0072] ;

[0073] In the formula, This represents the objective function value of the upper-level game model. The total number of monitoring areas is 4, which corresponds to 4 fan-shaped monitoring areas covered by 4 personnel detection cameras; For the first The predicted average thermal sensation index value for each monitoring area; The target is to predict the average thermal sensation index, with a value of 0; 0.5 is the standard thermal comfort deviation. For the first Population density index of each monitoring area, in person / 0.1 represents the standard population density, with units of people per unit area. The constraints of the upper-level game model include the air delivery angle of the electric side-discharge nozzle. The airflow range is -45 degrees to 45 degrees Celsius, and the air velocity at the electrically powered side-discharge nozzle is... The range is 2 to 8, and the unit is m / s, which is the air volume supplied by the bottom air outlet. The range is from 2000 to 10000, and the unit is... / h. The objective function of the lower-level game model is expressed as follows:

[0074] ;

[0075] In the formula, This represents the objective function value of the lower-level game model. Total air volume supplied, unit: / h; This is the static pressure loss, expressed in Pa, calculated from the duct resistance characteristic curve based on the total supply air volume. Confirmed; 10000 is the standard total supply air volume, unit is... / h; 250 is the standard static pressure loss, in Pa. The formula for calculating the air volume of the electrically powered side-supply nozzle is as follows:

[0076] ;

[0077] In the formula, This refers to the air volume of the electrically powered side-flow nozzle, in units of... / h; The air velocity is measured in m / s from the electrically powered side-flow nozzle. This refers to the cross-sectional area of ​​the electrically driven side-feed nozzle, in units of... The empirical value is 0.12; 3600 is the time conversion factor, in seconds per hour. Total air supply volume. The calculation formula is expressed as follows:

[0078] ;

[0079] In the formula, This refers to the air volume supplied by the bottom air outlet, in units of... The constraints of the lower-level game model include that the total supply air volume must meet the requirement of an indoor air exchange rate of not less than 6 times / h and the air velocity in each monitoring area must not exceed 0.5 m / s. The coupling terms of the two-level game model for air supply regulation are expressed as follows:

[0080] ;

[0081] In the formula, For coupling terms; Volume in three-dimensional space, unit: ; This represents the temperature gradient, expressed in °C / m. The velocity gradient is expressed in units of 1000 ppm. The first 1 in the denominator represents the standard temperature gradient, in °C / m, with an empirical value of 1; the second 1 in the denominator represents the standard velocity gradient, in °C / m. The empirical value is 1. The Nash equilibrium method is used to solve the two-level game model of air supply regulation, seeking the combination of air supply parameters that simultaneously achieves the relatively optimal objective functions of both the upper and lower level game models, including the air supply angle of the electrically driven side nozzles. Electric side-discharge nozzle airflow speed and the air volume of the bottom air outlet .

[0082] The specific implementation of step S06 is as follows: the heat exchange calculation formula is expressed as follows:

[0083] ;

[0084] In the formula, Heat exchange is measured in W. Air density, unit: kg / m³ Under standard conditions, its value is 1.2; The specific heat capacity of air at constant pressure is expressed in J / (kg·K), and its value is 1005. Total air volume supplied, unit: / h; This refers to the return air temperature, expressed in °C. The target outlet air temperature of the heat exchanger is given in °C; 3600 is the time conversion factor in seconds per hour. The formula for calculating the heat transfer temperature difference entropy production rate is as follows:

[0085] ;

[0086] In the formula, The heat transfer temperature difference entropy production rate, in W / K; Heat exchange is measured in W. This is the refrigerant-side temperature, in K, determined based on the refrigerant flow rate and heat exchanger characteristics. This refers to the air-side temperature, expressed in Kelvin (K), with values ​​ranging from [value missing]. 273.15 is the standard temperature, in Kelvin. The formula for calculating the entropy production rate of flow resistance is as follows:

[0087] ;

[0088] In the formula, The entropy production rate of flow resistance is expressed in W / K. Volumetric flow rate, unit: / s, with a value of ; Pressure loss, measured in Pa, is determined based on fluid velocity and pipe characteristics. The fluid temperature is expressed in Kelvin (K), and its value ranges from [value missing]. 101325 is the standard pressure, in Pa. The objective function of the heat transfer optimization model is expressed as follows:

[0089] ;

[0090] In the formula, The objective function value for the heat exchanger optimization model is given in W / K. The Kuhn-Tack condition equations are solved iteratively using the gradient descent algorithm to obtain the optimal refrigerant flow rate for the heat exchanger. and the target outlet air temperature of the heat exchanger ,in The unit is kg / s, and the range is 50% to 100% of the rated flow rate; The unit is ℃, and it is not lower than the dew point temperature.

[0091] The specific implementation of step S08 is as follows: the formula for calculating the deviation value of the predicted average thermal sensation index is expressed as follows:

[0092] ;

[0093] In the formula, To predict the deviation value of the average thermal sensation index; This represents the actual predicted average thermal sensation index value. The target is to predict the average thermal sensation index, with a value of 0. When... At that time, the predicted average thermal sensation index deviation value will be... Return to step S04 as feedback correction parameters; when When, return to step S01; where, The comfort deviation threshold is dynamically set based on the activity type and clothing of people indoors. The default value is 0.3 for office spaces, 0.5 for commercial spaces, and 0.7 for industrial spaces.

[0094] The implementation of the position-encoded sequential position-aware framework in the airflow field reconstruction model is as follows, and the position encoding formula is expressed as follows:

[0095] ;

[0096] ;

[0097] In the formula, coordinates The One coded component; coordinates The One coded component; This is the encoding channel index, with values ​​of 0, 1, and 2. For spatial nodes Coordinate components, in meters; This is a frequency parameter, in units of... The values ​​are 1, 10, and 100, respectively. coordinates and The coordinates are encoded in the same way, and the position codes of the three coordinate components are concatenated to form an 18-dimensional position feature vector. The 18-dimensional positional feature vector The physical features of the spatial nodes, including initial temperature and initial wind speed, are concatenated and used as input to the graph convolutional layer. In the graph convolution operation, the spatial node features are aggregated by weighted adjacency matrix to aggregate the information of adjacent spatial nodes. The position encoding enables the spatial flow field prediction network to distinguish the influence weight of spatial nodes at different spatial locations on the current spatial node during the aggregation process. Larger aggregation weights are assigned to spatial nodes that are closer, and smaller aggregation weights are assigned to spatial nodes that are farther away.

[0098] The formula for calculating the aggregation weight coefficient of the convolutional layer in the airflow field reconstruction model is as follows:

[0099] ;

[0100] In the formula, For spatial nodes To space node Aggregate weight coefficient; For spatial nodes Population density index of the monitored area, in person / 0.1 represents the standard population density, with units of people per unit area. ; This refers to the air volume supplied by the bottom air outlet, in units of... / h; 10000 is the standard supply air volume, unit is / h; For spatial nodes With spatial nodes The distance between them is in meters (m); 1 is the standard distance in meters; and 2 is the distance decay index. The aggregation weighting coefficient enables the airflow field reconstruction model to have stronger feature aggregation capabilities for spatial nodes in monitoring areas with dense populations and large airflow from the downdraft vents, thereby enhancing the prediction accuracy of airflow diffusion.

[0101] To better understand and implement this invention, the following is a specific application scenario of the invention, Example 2: The technical team deployed distributed air handling equipment in the central lobby of a large office building, which has an area of ​​approximately 500 square meters. The hall has a ceiling height of 4.5 meters and accommodates 80 to 150 people daily. A distributed air handling unit is installed in the center of the hall. Four people detection cameras face east, south, west, and north, covering a radius of 15 meters. Thermal comfort sensors are installed at 18 seats to collect environmental parameters. After the system is activated, the people detection cameras capture images of the people distribution at 9:00 AM, such as... Figure 2 As shown, there were 92 people in the hall during this period, mainly concentrated in the east and south areas. Environmental parameters collected by the thermal comfort sensor included a dry-bulb temperature of 26.8℃, relative humidity of 58%, mean radiant temperature of 27.2℃, and air velocity of 0.18 m / s.

[0102] The airflow field reconstruction model discretizes the three-dimensional space surrounding the device into 2400 spatial nodes with a node spacing of 0.5m, constructing a spatial topology graph structure containing 9600 directed edges. For example... Figure 3 As shown, the three-dimensional coordinates of the spatial nodes are embedded into the 128-dimensional feature space through sine and cosine position encoding, with the x-coordinate using frequencies of 1, 10, and 100. The sine and cosine functions generate six encoding channels. The y-coordinate and z-coordinate are encoded in the same way, and the three coordinate components are concatenated to form an 18-dimensional positional feature vector. The personnel distribution image is input into the airflow field reconstruction model. The model is processed through three graph convolutional layers and two fully connected layers. The first graph convolutional layer maps the 128-dimensional input to 256-dimensional features, the second layer maps it to 512-dimensional features, and the third layer regresses it to 256-dimensional features. The two fully connected layers sequentially output 128-dimensional features and the final four-channel flow field data. The model's forward propagation time is 12 milliseconds, and the output three-dimensional spatial flow field distribution data covers 2400 spatial nodes, including a temperature field distribution ranging from 24.5℃ to 28.3℃, and a velocity field distribution with a horizontal component ranging from -0.32 m / s to 0.45 m / s and a vertical component ranging from -0.15 m / s to 0.28 m / s.

[0103] The hall was divided into four sector-shaped monitoring areas, each with a central angle of 90 degrees and a radius of 15 meters. For example... Figure 4 As shown, 38 people were detected in the eastern monitoring area, covering an area of ​​176.7 square meters. The population density index is 0.215 people / 29 people were detected in the southern monitoring area, with a population density index of 0.164 people / day. Sixteen people were detected in the western monitoring area, with a population density index of 0.091 people / ... Nine people were detected in the monitoring area on the north side, with a population density index of 0.051 people / day. According to the ISO 7730 standard, taking into account the human metabolic rate of 1.2 MET, clothing thermal resistance of 0.7 clo, dry bulb temperature of 26.8℃, average radiant temperature of 27.2℃, air velocity of 0.18 m / s, and relative humidity of 58%, the predicted average thermal sensation index of the personnel activity area is calculated to be 0.62, indicating a slightly warm state. The target predicted average thermal sensation index is set to 0.

[0104] A two-layer game theory model for air supply regulation was established to solve for the optimal air supply parameters. The upper-layer game theory model aims to maximize comfort, using a weighted thermal comfort evaluation index that integrates the predicted average thermal sensation index deviation and population density weights for each monitoring area. The population density weight is 2.15 for the eastern monitoring area, 1.64 for the southern area, 0.91 for the western area, and 0.51 for the northern area. The standard thermal comfort deviation is set to 0.5, and the standard population density weight is set to 1. The lower-layer game theory model aims to minimize energy consumption, with the normalized air supply energy consumption function input including the total air supply volume and static pressure loss. The integral value of the product of the temperature gradient and velocity gradient in the three-dimensional spatial flow field distribution data is 78.6. This coupling term affects both upper-level comfort and lower-level energy consumption. Solving the two-level game model using the Nash equilibrium method yields an electric side-discharge nozzle airflow angle of -18 degrees, an electric side-discharge nozzle airflow velocity of 4.8 m / s, and a lower airflow volume of 6400 m³ / s. / h. The corresponding air volume for the electric side-discharge nozzle is 3200. / h, total air volume supplied is 9600 Based on the duct resistance characteristic curve, the static pressure loss is determined to be 186 Pa, and the normalized air supply energy consumption is 0.71, which meets the requirement of 8.5 air changes per hour. The air velocity in each monitoring area is below 0.5 m / s.

[0105] Based on the obtained air supply parameters, the intelligent control cabinet outputs control commands to adjust the electric side air supply nozzle actuators and the lower air supply outlet valves. Combining the temperature field distribution in the three-dimensional spatial flow field data, with an outdoor fresh air temperature of 32℃ and a return air temperature of 27.5℃, a heat transfer optimization model minimizing thermodynamic entropy production is established. The heat transfer temperature difference entropy production rate is caused by the temperature difference between the refrigerant side and the air side, while the flow resistance entropy production rate is caused by pressure loss. Constraints include a heat exchange capacity meeting an indoor heat load of 48kW, a refrigerant flow rate in the heat exchanger ranging from 50% to 100% of the rated flow rate of 1.8kg / s, and a target outlet air temperature in the heat exchanger not lower than the dew point temperature of 16.5℃. The temperature range of 15℃ to 35℃ is divided into 10 temperature sub-intervals, where the physical properties within each sub-interval are approximately linear functions. The constrained optimization problem was transformed into an unconstrained problem using the Lagrange multiplier method. The Kuhn-Tack condition equations were then solved iteratively using the gradient descent algorithm, with the iteration step size adaptively adjusted based on the eigenvalues ​​of the Hessian matrix. After 23 iterations, convergence was achieved, yielding an optimal refrigerant flow rate of 1.35 kg / s and a target outlet air temperature of 18.2°C. An intelligent control cabinet then controls the water flow regulating valve of the heat exchanger to adjust the flow rate to the target value.

[0106] Twenty-two seconds after the system adjusts the air supply, it re-acquires images of personnel distribution and environmental parameters. The thermal comfort sensor collects data showing a dry-bulb temperature of 25.9℃, relative humidity of 56%, mean radiant temperature of 26.3℃, and air velocity of 0.23 m / s, calculating an actual predicted average thermal sensation index of 0.28. The predicted average thermal sensation index deviation is 0.28, which is less than the office comfort deviation threshold of 0.3, meeting the comfort requirements. The system then returns to its initial state to continue monitoring personnel distribution. If the absolute value of the predicted average thermal sensation index deviation exceeds the comfort deviation threshold, this deviation is used as a feedback correction parameter. The weight coefficient of the predicted average thermal sensation index deviation is increased in the objective function of the upper-level game model, and the two-level game model for air supply adjustment is re-solved. Through multiple iterations, a closed-loop control system is formed.

[0107] The fresh air inlet is located at the bottom of the distributed air handling unit, with a 20-degree deflector. Outdoor fresh air is delivered into the room at a low speed of 0.3 m / s, mixing with the circulating air from the downflow vents near the ground to form a uniformly heated airflow layer. The concentration monitoring value was 920 ppm, which is within the normal range of 800 ppm to 1000 ppm. The outdoor fresh air volume was maintained at 2400. / h. The intelligent control cabinet is linked with the building's fire protection system to monitor the indoor temperature in real time. When the temperature exceeds 60°C, it is determined that a fire has occurred. The return air duct is immediately closed, the smoke exhaust mode is activated, and an alarm message is sent to the building management system.

[0108] As shown in Table 1, the airflow field reconstruction model has different prediction accuracy in different monitoring areas. The prediction accuracy is higher in the monitoring areas on the east and south sides with higher population density. This is because the aggregation weight coefficient is dynamically adjusted according to the population density index and the air volume supplied by the lower air outlet, which gives the spatial nodes a stronger feature aggregation capability in densely populated areas.

[0109] Table 1. Flow field prediction accuracy in different monitoring areas

[0110]

[0111] The technological advancements of this invention compared to traditional air conditioning control methods are reflected in three aspects. First, the airflow field reconstruction model establishes a spatial topology structure based on a graph neural network. By mapping the three-dimensional coordinates of spatial nodes into high-dimensional feature vectors through position encoding, the model can capture the spatial relative relationships and anisotropic characteristics of airflow propagation. This avoids the complex iterative process of solving the Navier-Stokes equations using traditional computational fluid dynamics methods, reducing the time required for a single flow field reconstruction from several minutes to milliseconds, meeting the response speed requirements of real-time control. Second, the two-layer game theory model for air supply regulation simultaneously optimizes both comfort and energy consumption objectives using the Nash equilibrium method. The upper-layer game theory model assigns differentiated comfort weights to each monitoring area based on the population density index, while the lower-layer game theory model balances airflow mixing intensity and turbulent dissipation through the coupling term of temperature gradient and velocity gradient. Compared to traditional single-objective optimization methods, this approach is more in line with practical application needs. Finally, the heat exchange optimization algorithm that minimizes thermodynamic entropy production establishes an irreversible loss model based on the second law of thermodynamics. It comprehensively considers the entropy production of heat transfer temperature difference and the entropy production of flow resistance, and solves for the optimal heat exchanger parameters through the Lagrange multiplier method and gradient descent algorithm. This avoids the limitations of traditional trial-and-error methods or empirical formulas and improves the thermodynamic efficiency of the heat exchange system.

[0112] It should be noted that the variables involved in this invention are explained in detail in Table 2.

[0113] Table 2 Variable Explanation Table

[0114]

[0115] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart adjustment method for a distributed air handling unit, characterized in that, After the distributed air handling unit starts operating, personnel distribution images of the surrounding area are collected by personnel detection cameras, and dry-bulb temperature, relative humidity, mean radiant temperature, and air velocity data of the personnel activity area are collected by thermal comfort sensors. The personnel distribution images are input into the airflow field reconstruction model to output three-dimensional spatial flow field distribution data. The personnel density index of each monitoring area is calculated based on the personnel distribution images, and the predicted average thermal sensation index of the personnel activity area is calculated based on the dry-bulb temperature, relative humidity, mean radiant temperature, and air velocity data. Based on the personnel density index, the predicted average thermal sensation index, and the three-dimensional spatial flow field distribution data, a two-layer game model for air supply regulation is established to solve for the air supply angle, air velocity, and air volume of the lower air supply outlet, and to control the electric air supply. The side-supply nozzle actuator adjusts the air supply angle and air supply speed, and controls the air supply valve at the lower air supply outlet to adjust the air supply volume. Based on the temperature field distribution and the outdoor fresh air temperature and return air temperature, the heat exchanger target outlet air temperature and refrigerant flow rate are calculated using a heat exchange optimization algorithm that minimizes thermodynamic entropy production. The heat exchanger water flow regulating valve is controlled to adjust the flow rate. The personnel distribution image, dry-bulb temperature, relative humidity, mean radiant temperature, and air velocity data are re-acquired to calculate the actual predicted average thermal sensation index value. The predicted average thermal sensation index deviation value between the actual predicted average thermal sensation index value and the target predicted average thermal sensation index value is calculated. When the absolute value of the predicted average thermal sensation index deviation value is greater than the comfort deviation threshold, the predicted average thermal sensation index deviation value is used as a feedback correction parameter and returned to the air supply regulation two-layer game model for re-solving.

2. The method according to claim 1, characterized in that, There are four personnel detection cameras, which are installed in the four directions of the distributed air handling equipment to collect information on the distribution of people within a 15m radius around the equipment from all directions.

3. The method according to claim 2, characterized in that, The structure of the airflow field reconstruction model is a spatial flow field prediction network based on graph neural network. The three-dimensional space covered by the distributed air handling equipment is discretized into several spatial nodes. The airflow transmission relationship between adjacent spatial nodes is represented by directed edges to construct a spatial topology graph structure. The coordinates of the spatial nodes are embedded into the high-dimensional feature space through sine and cosine position encoding.

4. The method according to claim 3, characterized in that, The spatial flow field prediction network consists of three graph convolutional layers and two fully connected layers. The first graph convolutional layer has an input dimension of 128 and an output dimension of 256. The second graph convolutional layer has an input dimension of 256 and an output dimension of 512. The third graph convolutional layer has an input dimension of 512 and an output dimension of 256. The two fully connected layers map the 256-dimensional features to 128 dimensions and the final output dimension, respectively. The output dimension includes four channels: temperature component and velocity components in three directions. Residual connections are used to add the input features to the outputs of each layer.

5. The method according to claim 4, characterized in that, Sine and cosine position encoding maps the three-dimensional coordinates of spatial nodes into high-dimensional vectors through sine and cosine functions of different frequencies. This enables the spatial flow field prediction network to automatically perceive the spatial positional relationship of spatial nodes when learning the airflow transmission law, avoiding feature confusion caused by spatial translation.

6. The method according to claim 5, characterized in that, The steps for establishing the training dataset for the airflow field reconstruction model include building a distributed air handling equipment model with a scale of 1:5 in a laboratory environment, arranging 200 temperature sensors and 200 wind speed sensors around the distributed air handling equipment model to form a dense grid of measuring points, collecting time series data of temperature and wind speed at each measuring point under different air supply conditions, and collecting a total of 500 sets of flow field data under different air supply conditions by changing the air supply angle, air supply speed, personnel distribution pattern and obstacle position.

7. The method according to claim 6, characterized in that, The training steps of the airflow field reconstruction model use mean squared error as the loss function and the Adam optimizer for parameter updates. The initial learning rate is set to 0.001, and the learning rate is reduced to 0.5 times every 30 training cycles. The batch size is set to 32, and the total number of training cycles is 200. Training is stopped early when the loss function value of the validation set no longer decreases after 20 consecutive training cycles.

8. The method according to claim 7, characterized in that, The method for calculating the population density index is to divide the personnel distribution image captured by the personnel detection camera into several monitoring sub-regions. Each monitoring sub-region corresponds to a fan-shaped area around the device. The central angle of the fan-shaped area is 90 degrees and the radius is 15m. Human detection is performed on the image of each monitoring sub-region to identify the number of people in the monitoring sub-region. The number of people is divided by the area of ​​the monitoring sub-region to obtain the population density index.

9. The method according to claim 8, characterized in that, The predicted average thermal sensation index is the PMV value calculated according to the international standard ISO7730. The calculation formula takes into account six parameters: human metabolic rate, clothing thermal resistance, dry bulb temperature, mean radiant temperature, air velocity, and relative humidity. The dimensionless value characterizing human body's hot and cold sensation is obtained by solving the human body thermal balance equation.

10. The method according to claim 9, characterized in that, The two-layer game model for air supply regulation includes an upper-layer game model that aims to maximize comfort and a lower-layer game model that aims to minimize energy consumption. The objective function of the upper-layer game model is to maximize the weighted thermal comfort function, while the objective function of the lower-layer game model is to minimize the air supply energy consumption function. The two objective functions are related through a coupling term, which is the integral of the product of the temperature gradient and velocity gradient of each spatial node in the three-dimensional spatial flow field distribution data in space. The two-layer game model for air supply regulation is solved using the Nash equilibrium method.

Images