High-precision variable air volume valve

Abstract

The invention relates to a high-precision variable air volume valve and an air volume control method thereof. The variable air volume valve is composed of a valve body, a front valve block set, a rear valve block, two actuators and a controller, linear adjustment of the air volume can be achieved, the air volume adjustment precision is improved, and adjustment of high and low air volume intervals is supported. The controller collects pressure difference data of an inlet and an outlet through the pressure difference sensor, and the real-time air volume is calculated through a built-in Kolmogorov-Arnold (KAN) neural network model. By comparing the actual air volume with the target air volume in real time, the controller adjusts the actuator to change the opening degree of the front valve plate set and the opening degree of the rear valve plate set, and accurate air volume control is achieved. According to the technical scheme, the accuracy and the adjusting range of air volume measurement are improved, additional resistance in the air pipe is reduced, and the performance and the operation efficiency of an air conditioning system are remarkably optimized.

Description

Technical Field

[0001] The invention relates to the technical field of design, manufacture and control of air valves in ventilation and air-conditioning engineering, and in particular to a VAV air valve. Background Art

[0002] In modern large-scale public buildings and industrial buildings, the operation and regulation of air conditioning systems are key factors in ensuring indoor environmental comfort and air quality. Variable Air Volume (VAV) valves are an indispensable component in air conditioning systems. By adjusting the air volume in the air duct, precise control of regional temperature, humidity and cleanliness is achieved. Traditional variable air volume valves are usually composed of a valve body, a valve plate, an actuator, a pressure measuring tube, a differential pressure sensor and a controller. The opening of the valve plate is adjusted by an electric actuator to change the resistance in the air duct, thereby controlling the air volume. However, the existing variable air volume valves have the following major problems in terms of air volume acquisition and adjustment accuracy:

[0003] 1. Insufficient air volume measurement accuracy: Conventional variable air volume valves usually use a pitot tube pressure gauge to measure the total pressure and static pressure of the air, and then calculate the real-time air volume through the Bernoulli equation. However, the pitot tube coefficients of different forms of pressure gauges and pressure-taking surfaces are difficult to obtain accurately, resulting in errors in calculating the air volume. In addition, the measurement range of this method is limited, and it can usually only meet the measurement range of about 30% to 100% of the rated air volume of the air valve. When the actual air volume is less than 30%, the measurement accuracy drops significantly, and it is difficult to guarantee accuracy.

[0004] 2. Limited air volume adjustment range: Traditional variable air volume valves mostly use a single actuator and a single valve plate design. This design can basically meet the needs of ordinary air volume adjustment, but it is difficult to improve the control accuracy when a large range of air volume adjustment is required. Especially when high and low air volume ranges are precisely adjusted at the same time, the single valve plate design seems to be unable to provide sufficient adjustment capability and control accuracy.

[0005] 3. Increased resistance in the air duct: Conventional pitot tube pressure measuring devices add additional resistance inside the air duct, affecting the efficiency and performance of the air system. Especially when the air duct size is large or the system is complex, the installation and maintenance of the pitot tube also brings additional difficulties.

[0006] In order to solve the above problems, further improving the air volume acquisition and control accuracy of the variable air volume valve, expanding its air volume control range, and reducing the additional resistance in the air duct have become important directions for the optimization of variable air volume valve technology. Achieving these goals is of great significance to improving the overall performance and operating efficiency of the air conditioning system. Summary of the invention

[0007] The purpose of the present invention is to provide a variable air volume valve that can accurately measure and control a wide range of air volume. The variable air volume valve adopts a two-position valve plate design and a neural network model to calculate the real-time air volume, improves the accuracy of air volume acquisition and adjustment, expands the air volume control range, and reduces the additional resistance in the air duct, thereby optimizing the performance of the variable air volume valve and meeting the needs of complex air conditioning systems in large public buildings and industrial buildings.

[0008] To achieve the above object, the present invention adopts the following technical solution:

[0009] Variable air volume valve structure: The variable air volume valve of the present invention is composed of a valve body, a front valve plate group, a rear valve plate, an actuator 1, an actuator 2, a controller, a front pressure measuring hole group, a front pressure measuring tube, a rear pressure measuring hole group and a rear pressure measuring tube. The inlet and outlet pressure difference of the air valve is transmitted to the pressure difference sensor in the controller through the front and rear pressure measuring hole groups and the pressure measuring tube. The MCU in the controller calculates the real-time air volume through the built-in neural network algorithm module, and adjusts the actuator according to the set target air volume to control the valve plate opening and realize air volume regulation.

[0010] Furthermore, the real-time air volume calculation method includes: by testing the front and rear valve openings and the air valve inlet and outlet pressure difference under different working conditions, the air volume calculation model is trained using the Kolmogorov-Arnold (KAN) neural network model. The input feature vector of the model includes the front valve opening, the rear valve opening and the air valve pressure difference, and the output is the predicted air volume value. After the model training is completed, the model is sunk into the MCU of the controller to calculate the air volume based on real-time data.

[0011] Furthermore, the controller structure and functions include: the controller is mainly composed of a board, an MCU, a pressure difference sensor, a WiFi module, a wired network port, a terminal group and other related circuit board components. The MCU is used to deploy the air volume calculation model and the control logic program; the pressure difference sensor reads the real-time pressure difference; and the controller transmits the air valve operation parameters through WiFi and wired networks.

[0012] Furthermore, the air valve pressure difference acquisition method includes: evenly opening pressure measuring holes on the upper and lower walls of the inlet and outlet ends of the air valve, transmitting the pressure difference to the pressure difference sensor in the controller through the pressure measuring tube for real-time air volume calculation. Specifically, symmetrical pressure measuring hole groups are respectively opened at the inlet and outlet ends of the air valve, and the number of pressure measuring holes is adjusted according to the size of the air valve, and is at least 8.

[0013] Furthermore, the front and rear valve plate design includes: the front valve plate group is arranged at the inlet end of the air valve, consisting of a single or two parallel valve plates, and the area is equal to the cross-sectional area of ​​the air valve; the rear valve plate is arranged at the outlet end of the air valve, and the area is 80% of the cross-sectional area of ​​the air valve. Two actuators control the opening of the front and rear valve plates respectively, and the precise adjustment of the air volume is achieved by coordinating the opening of the front and rear valve plates. The distance between the front valve plate group and the rear valve plate is designed to be slightly larger than half of the sum of the widths of the two to ensure that they do not interfere with each other when fully opened.

[0014] Furthermore, the air volume adjustment mechanism includes: the controller reads the deviation between the real-time air volume and the target air volume, first adjusts the opening of the front valve group to quickly approach the target air volume, and then achieves fine adjustment through fine adjustment of the rear valve. The front valve group completes 90%-95% of the adjustment task, and the rear valve completes 5%-10% of the fine adjustment, ensuring high-precision air volume control. This adjustment mechanism can achieve precise adjustment in both high and low air volume ranges.

[0015] Through the above-mentioned technical means, the present invention realizes accurate measurement and control of a wide air volume range, improves the accuracy and range of air volume regulation, reduces the additional resistance in the air duct, and significantly optimizes the performance of the variable air volume valve.

[0016] Furthermore, the process of using the Kolmogorov-Arnold (KAN) neural network model to train the air volume calculation model includes the following steps:

[0017] 1. Data testing and collection:

[0018] Determine the standard air volume Q std : Determine the standard air volume when the front and rear valves of the air valve are fully opened. Record initial data: Install the air valve on the air system test bench and adjust the air supply volume of the test device to Q std , keep the front and rear valves of the air valve fully open (opening degree is 90 degrees), record the opening degree of the front and rear valves of the air valve, the pressure difference between the inlet and outlet of the air valve and the air volume read by the air volume hood.

[0019] Adjust the opening of the rear valve: keep the opening of the front valve unchanged, adjust the opening of the rear valve in intervals of 15 degrees, and record the air volume and the pressure difference between the inlet and outlet of the air valve at different openings (75 degrees, 60 degrees, 45 degrees, 30 degrees, 15 degrees, and 0 degrees).

[0020] Adjust the front valve opening: adjust the front valve opening to 75 degrees, 60 degrees, 45 degrees, 30 degrees, 15 degrees, and 5 degrees respectively, repeat the above steps, and record the data under different working conditions.

[0021] 2. Model construction and training:

[0022] Summarize data: Summarize all the data obtained from the test to form a complete data set.

[0023] Data partitioning: The dataset is divided into a training set and a test set in a ratio of 8:2 for model training and verification.

[0024] Network structure: A Kolmogorov-Arnold (KAN) network model was constructed, which includes an input layer, three hidden layers, and an output layer. The number of nodes in each hidden layer is 200. The network is designed with a learnable activation function, and a B-spline function is set at the edge of the node to directly connect two neurons to replace the linear weight in the MLP. SiLU (Sigmoid Linear Unit) is used as the basic activation function.

[0025] Input feature vector: including front valve opening, rear valve opening and air valve pressure difference.

[0026] Grid parameters: Use a grid size of 5x5 and a third-order spline function, a grid range of (-1,1), and an accuracy of 0.02.

[0027] Noise and scaling factor: The ratio of adding noise is 0.1 to improve the generalization ability and robustness of the model, and the base scaling factor and spline scaling factor are both 1.0.

[0028] Optimizer selection: AdamW optimizer is used for model training. AdamW combines the advantages of Adam optimizer and L2 regularization effect, which helps the model converge quickly.

[0029] Loss function: The L2 loss function (mean square error, MSE) is used to measure the difference between the model prediction value and the actual value.

[0030] Model training: Use the training set data to train the KAN network model, and optimize the model's depth, width, and training batch size through automatic tuning.

[0031] 3. Model verification and deployment:

[0032] Model validation: Use the test set data to validate the performance of the trained KAN network model to ensure that the model has good prediction accuracy and generalization ability.

[0033] Model deployment: The trained KAN network model is deployed in the controller’s MCU to calculate the air volume of the air valve in real time.

[0034] 4. Real-time air volume calculation:

[0035] Real-time data input: The controller reads the front and rear valve openings and the air valve inlet and outlet pressure difference data in real time.

[0036] Air volume prediction: The KAN network model deployed in the MCU calculates the predicted air volume based on real-time data and displays it on the controller.

[0037] Through the above steps, the Kolmogorov-Arnold (KAN) neural network model is used to train the air volume calculation model, which can realize the real-time accurate calculation of the air volume of the air valve under different working conditions and improve the air volume adjustment accuracy and control range of the variable air volume valve.

[0038] The technical solution of the present invention has the following beneficial effects:

[0039] The present invention uses the Kolmogorov-Arnold (KAN) neural network model to replace the traditional pitot tube pressure measurement and Bernoulli equation calculation method, and can accurately calculate the real-time air volume. This method overcomes the error caused by the inaccurate pitot tube coefficient in the traditional measurement method, and can maintain high-precision air volume acquisition in the small air volume range.

[0040] Expanded air volume adjustment range: The present invention adopts a two-position valve plate design, through the front and rear two independent valve plate groups are controlled by two actuators respectively, to achieve linear adjustment of air volume. The front valve plate group is responsible for most of the air volume adjustment, and the rear valve plate is responsible for fine adjustment, ensuring accurate control in both high and low air volume ranges, meeting a wider range of air volume adjustment needs.

[0041] Reduce the resistance in the air duct: By evenly setting pressure measuring holes on the side wall at the inlet and outlet of the air valve, replacing the traditional pitot tube pressure measuring device, the additional resistance in the air duct is reduced and the overall efficiency of the air system is improved. At the same time, the pressure measuring hole design simplifies the installation and maintenance process and reduces operating costs.

[0042] Improve system operation efficiency: The controller of the present invention realizes accurate calculation and adjustment of real-time air volume through the built-in neural network model and high-precision differential pressure sensor. The controller can quickly respond and adjust the valve opening to ensure the accuracy and stability of air volume regulation, significantly improving the operation efficiency of the entire air conditioning system.

[0043] Intelligent control and communication: The controller of the present invention is equipped with a WiFi module and a wired network port, which can transmit the operating parameters of the air valve through a wireless or wired network and support remote monitoring and management. The controller adopts advanced MCU and optimized algorithm model to realize intelligent air volume adjustment control, enhancing the operability and convenience of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a schematic diagram of the valve body structure of a specific embodiment of the present invention;

[0045] Figure 2 It is a schematic diagram of a controller according to a specific embodiment of the present invention. DETAILED DESCRIPTION

[0046] The technical solution of the present invention is further explained below in conjunction with the accompanying drawings and specific embodiments.

[0047] The variable air volume valve of the present invention comprises a valve body (1), a front valve plate group (2), a rear valve plate (3), an actuator 1 (4), an actuator 2 (5), a controller (6), a front pressure measuring hole group (7), a front pressure measuring tube (8), a rear pressure measuring hole group (9) and a rear pressure measuring tube (10). The front pressure measuring hole group (7), the front pressure measuring tube (8), the rear pressure measuring hole group (9) and the rear pressure measuring tube (10) are used in combination to transmit the pressure difference through the inlet and outlet of the air valve to the pressure difference sensor 63 in the controller (6), and obtain the real-time air volume by relying on the neural network algorithm module built in the MCU (62) of the controller (6); the controller (6) compares the set target air volume with the real-time air volume, sends a command to the actuator 1 (4) and the actuator 2 (5), and realizes air volume regulation by adjusting the opening of the front valve plate group (2) and the rear valve plate (3).

[0048] Furthermore, the real-time air volume calculation method is:

[0049] 1. Data testing and collection:

[0050] (1) Determine the standard air volume Q when the front and rear valves of the air valve are fully open std , with a cross-section width of 320mm, a height of 250mm, and a Q std =1300m 3 Take the air valve with a capacity of / h as an example.

[0051] (2) Install the air valve on the air system test bench and adjust the air supply volume of the test device to Q std ; Keep the opening of the front valve plate and the rear valve plate of the air valve at 90 degrees (fully open), record the opening of the front valve plate, the opening of the rear valve plate, the pressure difference between the inlet and outlet of the air valve, and the air volume read by the air volume hood.

[0052] (3) Keep the opening of the front valve plate of the air valve unchanged, adjust the opening of the rear valve plate at intervals of 15 degrees, and record the air volume and the pressure difference between the inlet and outlet of the air valve when the rear valve opening is 75 degrees, 60 degrees, 45 degrees, 30 degrees, 15 degrees, and 0 degrees.

[0053] (4) Then adjust the front valve opening to 75 degrees, 60 degrees, 45 degrees, 30 degrees, 15 degrees, and 5 degrees respectively, and repeat step 3. The test conditions and data are shown in the following table:

[0054] Table 1Q std Test data under working conditions

[0055]

[0056]

[0057]

[0058] (5) Adjust the working conditions of the air system test bench and adjust the standard air volume to 0.25Q when the front and rear valves of the air valve are fully opened std , 0.5Q std , 0.75Q std , 1.25Q std , 1.5Q std , 1.75Q std , repeat steps (2) to (4) and record the test data under different working conditions (the remaining 6 groups of working conditions and test data are not listed).

[0059] 2. The model construction and training process is as follows:

[0060] The air volume calculation model uses the Kolmogorov-Arnold (KAN) network model to predict the air volume by inputting a feature vector. The input feature vector includes the front valve opening, the rear valve opening, and the air valve pressure difference. The output of the model is a single-element vector containing the predicted air volume value. The model training process is as follows:

[0061] (1) Data preprocessing: All the data obtained from the air valve test are summarized and evenly sampled and divided into training set and test set in a ratio of 8:2.

[0062] (2) Model construction, characterized in that:

[0063] (2.1) Network structure: It includes an input layer, three hidden layers and an output layer, where the number of nodes in the hidden layers is 200, 200, and 200 respectively. Linear weights are not included at all, and each weight parameter is replaced by a parameterized spline function.

[0064] (2.2) Grid parameters: Use a grid size of 5x5 and a third-order spline function.

[0065] (2.3) Noise and scaling factor: The ratio of adding noise is 0.1, which is used to improve the generalization ability and robustness of the model. The base scaling factor is 1.0, and the spline scaling factor is also 1.0, which is used to adjust the overall weight of the network and the influence of the spline function.

[0066] (2.4) Activation function: Different from traditional MLP neural networks, KAN uses a learnable activation function design on the node edge (weight) rather than within the node, so as to achieve dynamic adjustment according to data and tasks. SiLU (Sigmoid Linear Unit) is used as the basic activation function.

[0067] (2.5) Grid range and precision: The grid range is (-1,1), which is used to define the area where the model samples in the input space. The grid precision is 0.02, which is used to control the sampling rate or resolution of the input data.

[0068] (3) Optimizer selection: The AdamW optimizer is used for model training. AdamW is an adaptive learning rate optimization algorithm that combines the advantages of the Adam optimizer and the effect of L2 regularization. In each iteration, AdamW updates the weights according to the gradient of the model and automatically adjusts the learning rate to help the model converge quickly.

[0069] (4) Automatic tuning: During the training process, the depth and width of the model, as well as the batch size of the training, are automatically tuned.

[0070] (5) Loss function: The L2 loss function is used to measure the difference between the model prediction value and the actual value. The L2 loss function, also known as the mean square error (MSE), is a commonly used loss function for regression problems. It calculates the sum of the squares of the difference between the predicted value and the actual value to guide the model training process.

[0071] The entire model program framework is converted and written in C language. After the model training is completed, it is laid down in the MCU (62) of the controller (6). Based on the laid-out model, the relationship between the front valve plate opening, the rear valve plate opening, the air valve inlet and outlet pressure difference and the air volume is learned from the test data, so that the real-time air volume can be automatically calculated (predicted) and displayed according to the real-time data related to the air valve when the air system is running.

[0072] In this embodiment, the specific structure and function of the controller of the variable air volume valve are as follows: the controller (6) is mainly composed of a board (61), an MCU (62), a pressure difference sensor (63), a WiFi module (64), a wired network port (65), a terminal group 1 (66), a terminal group 2 (67), a terminal group 3 (68), and other related circuit board components. Among them, the MCU (62) is an embedded core single chip microcomputer, which is used to deploy a neural network model for air volume calculation and a logic program for controlling air volume; the performance requirements of the MCU used in the present invention are: 1) the number of pins is at least 80; 2) the number of ADC pins provided is at least 6; 3) the flash memory is at least 2Mbytes; 4) the size of the RAM is at least 512kbyte.

[0073] In this embodiment, the differential pressure sensor (63) is used to read the real-time differential pressure between the inlet and outlet ends and transmit it to the MCU (62); the range of the differential pressure sensor (63) is 0-500Pa, and the measurement accuracy is ±1Pa. Terminal group 1 (66) includes the motor interface of the two actuators and the potentiometer interface for feedback of the opening, terminal group 2 (67) includes the power interface and the RS485 communication interface, and terminal group 3 (68) is mainly an analog interface, including the potentiometer signal port (opening) of the two actuators, the differential pressure signal port and the air volume signal port of the differential pressure sensor (63), which can transmit the opening of the front and rear valve plates, the real-time differential pressure and the air volume data to the outside in the form of current and voltage; at the same time, the operating parameters of the air valve, including the opening of the front and rear air valves, the real-time differential pressure and the air volume can also be transmitted to the outside through the WiFi module (64) and the wired network port (65).

[0074] In this embodiment, the method for obtaining the inlet and outlet pressure difference of the air valve and the related structural design are as follows: at the inlet end of the air valve, N 10 mm pressure measuring holes are evenly and symmetrically opened on the upper and lower walls. The number of pressure measuring holes is adjusted according to the size of the air valve. The larger the size of the air valve, the more pressure measuring holes can be provided, but at least 8. These pressure measuring holes form a front pressure measuring hole group (7). A Y-shaped front pressure measuring tube (8) is used to connect the pressure measuring holes on the upper and lower walls to obtain the pressure on the inlet end face. The outlet pipe of the front pressure measuring tube (8) is connected to the pressure differential sensor (63) in the controller (6). Similarly, at the outlet end of the air valve, N 10 mm pressure measuring holes are evenly and symmetrically opened on the upper and lower walls. These pressure measuring holes form a rear pressure measuring hole group (9). A Y-shaped rear pressure measuring tube (10) is used to connect the pressure measuring holes on the upper and lower walls to obtain the pressure on the outlet end face. The outlet pipe of the rear pressure measuring tube (9) is connected to the low-pressure inlet of the pressure differential sensor (63) in the controller (6). The pressure differential obtained by the pressure differential sensor (63) in the controller (6) is the pressure differential at the inlet and outlet ends of the air valve when it is working, which is used to calculate and obtain the real-time air volume of the air valve.

[0075] In this embodiment, the front valve plate group and the rear valve plate in the variable air volume valve are designed as follows: the front valve plate group (2) is arranged at the inlet end of the air valve, and its composition can be a single valve plate or a valve plate group composed of two parallel valve plates. The area of ​​the front valve plate or the valve plate group is equal to the cross-sectional area through which the air valve passes. When the front valve plate group (2) is fully closed, the air valve is in a closed state. The actuator 1 (4) is fixed on the central axis of the front valve plate group (2). If there are two parallel valve plates, the central axes of the two parallel valve plates will be connected by a linkage rod, and the actuator only needs to be fixed on one of the central axes. The actuator drives the central axis to rotate by rotating. The valve plate angle is then adjusted to achieve the purpose of controlling the opening of the front valve plate group (2); the rear valve plate (3) is arranged at the outlet end of the air valve, the rear valve plate (3) is a single valve plate, and the area of ​​the rear valve plate is 80% of the cross-sectional area of ​​the air valve; the actuator 2 (5) is fixed on the central axis of the rear valve plate (3), and the actuator drives the central axis to rotate by rotation to adjust the valve plate angle to achieve the purpose of controlling the opening of the rear valve plate; the distance between the central axes of the front valve plate group (2) and the rear valve plate (3) is slightly larger than 1 / 2 of the sum of the width of the front valve plate and the width of the rear valve plate, and the front valve plate group (2) and the rear valve plate (3) do not interfere with each other when they are fully opened.

[0076] In this embodiment, the air volume adjustment mechanism of the variable air volume valve is as follows: when the air system is running, the controller (6) reads the real-time air volume (air valve model, real-time pressure difference, real-time opening) calculated in the MCU (62), and compares it with the target air volume preset in the controller (6). First, the opening of the front valve plate group (2) is controlled and adjusted so that the real-time air volume quickly approaches the target air volume, and then the opening of the rear valve plate (3) is adjusted to achieve precise adjustment of the air volume. The front valve plate group (2) completes 90%-95% of the air volume adjustment task, and the rear valve plate (3) completes 5%-10% of the air volume fine adjustment. The front valve plate group (2) and the rear valve plate (3) are combined for adjustment, so that the air volume adjustment of the air valve is more linear, thereby improving the air volume adjustment accuracy. On the other hand, this method of rapid adjustment of the front valve plate (2) and fine adjustment of the rear valve plate can achieve precise adjustment in both high and low air volume ranges.

Claims

1. A high-precision variable air volume valve, comprising a valve body; a front valve plate group, arranged at the inlet end of the valve body; a rear valve plate, arranged at the outlet end of the valve body; an actuator 1, used to control the opening of the front valve plate group; an actuator 2, used to control the opening of the rear valve plate; a controller, comprising a pressure difference sensor, an MCU, a WiFi module, a wired network port, a front pressure measuring hole group, a front pressure measuring tube, a rear pressure measuring hole group and a rear pressure measuring tube, the front and rear pressure measuring hole groups and the pressure measuring tube transmit the inlet and outlet pressure difference of the air valve to the pressure difference sensor in the controller, the MCU in the controller calculates the real-time air volume through the built-in neural network algorithm module, and adjusts the actuator according to the set target air volume, controls the valve plate opening, and realizes air volume regulation; The model construction and training method of the neural network algorithm module are as follows: Summarize data: Summarize all the data obtained from the test to form a complete data set; Data partitioning: The data set is divided into training set and test set in a ratio of 8:2 for model training and verification; Network structure: Construct a Kolmogorov-Arnold network model, which includes an input layer, three hidden layers and an output layer. The number of nodes in each hidden layer is 200. The network adopts a learnable activation function design, uses SiLU as the basic activation function, and uses a B-spline function (B-spline) on the edge to directly connect two neurons to replace the linear weight in MLP; Input feature vector: including front valve opening, rear valve opening and air valve pressure difference; Grid parameters: a 5x5 grid size and a third-order spline function are used, the grid range is (-1,1), and the precision is 0.02; Noise and scaling factor: the ratio of adding noise is 0.1 to improve the generalization ability and robustness of the model, and the basic scaling factor and spline scaling factor are both 1.0; Optimizer selection: AdamW optimizer is used for model training. AdamW combines the advantages of Adam optimizer and L2 regularization effect, which helps the model converge quickly. Loss function: L2 loss function (mean square error, MSE) is used to measure the difference between the model prediction value and the actual value; Model training: Use the training set data to train the KAN network model, and optimize the model's depth, width, and training batch size through automatic tuning.

2. A high-precision variable air volume valve according to claim 1, characterized in that: The neural network algorithm module calculates the real-time air volume and further includes the following steps: data testing and collection: determining the standard air volume Q std : Determine the standard air volume when the front and rear valves of the air valve are fully opened; record the initial data: install the air valve on the air system test bench, and adjust the air supply volume of the test device to Q std , keep the front and rear valves of the air valve fully open, record the opening of the front and rear valves of the air valve, the pressure difference between the inlet and outlet of the air valve and the air volume read by the air volume hood; adjust the opening of the rear valve: keep the opening of the front valve unchanged, adjust the opening of the rear valve at intervals of 15 degrees, and record the air volume and the pressure difference between the inlet and outlet of the air valve at different openings; adjust the opening of the front valve: adjust the opening of the front valve to 75 degrees, 60 degrees, 45 degrees, 30 degrees, 15 degrees, and 5 degrees respectively, repeat the above steps, and record the data under different working conditions.

3. A high-precision variable air volume valve according to claim 2, characterized in that: The neural network algorithm module calculates the real-time air volume and also includes the following steps: model verification: using the test set data to verify the performance of the trained KAN network model to ensure that the model has good prediction accuracy and generalization ability; model deployment: sinking the trained KAN network model to the MCU of the controller for real-time calculation of the air volume of the air valve.

4. A high-precision variable air volume valve as claimed in claim 3, characterized in that: The neural network algorithm module calculates the real-time air volume and also includes the following steps: real-time data input: the controller reads the front and rear valve openings and the air valve inlet and outlet pressure difference data in real time; air volume prediction: through the KAN network model deployed in the MCU, the predicted air volume is calculated based on the real-time data and displayed on the controller.

5. A high-precision variable air volume valve as claimed in claim 4, characterized in that: The controller includes a board, MCU, a pressure difference sensor, a WiFi module, a wired network port, and a terminal group. The MCU is used to deploy the air volume calculation model and control logic program; the pressure difference sensor reads the real-time pressure difference; and the controller transmits the air valve operating parameters through WiFi and wired networks.

6. A high-precision variable air volume valve according to claim 1, characterized in that: The front valve plate group is arranged at the inlet end of the air valve, and is composed of a single or two parallel valve plates, with an area equal to the cross-sectional area of ​​the air valve; the rear valve plate is arranged at the outlet end of the air valve, and its area is 80% of the cross-sectional area of ​​the air valve. The two actuators control the opening of the front and rear valve plates respectively, and the precise adjustment of the air volume is achieved by coordinating the opening of the front and rear valve plates. The distance between the front valve plate group and the rear valve plate is designed to be greater than half of the sum of their widths to ensure that there is no mutual interference when fully open.

7. A high-precision variable air volume valve according to claim 1, characterized in that: Pressure measuring holes are evenly opened on the upper and lower walls of the inlet and outlet ends of the air valve. The pressure difference is transmitted to the pressure difference sensor in the controller through the pressure measuring tube for real-time air volume calculation.

8. A high-precision variable air volume valve according to claim 1, characterized in that: When the air system is running, the controller reads the real-time air volume calculated in the MCU and compares it with the target air volume preset in the controller. First, it controls the opening of the front valve group to make the real-time air volume quickly approach the target air volume, and then adjusts the opening of the rear valve to achieve precise adjustment of the air volume. The front valve group completes 90%-95% of the air volume adjustment task, and the rear valve completes 5%-10% of the fine adjustment of the air volume.

Images