Multi-zone distributed intelligent regulation and control method, system and equipment for clean room and medium

Abstract

The method comprises the steps that the number, positions and activity types of people in the clean room are obtained in real time, the intensity of particulate matter emission sources of people in all areas is determined, and the particulate matter concentration of all the areas is calculated through a particulate matter concentration prediction model in combination with the current air supply amount of all the areas; and under the constraint that the particle concentration of each area does not exceed the standard, with the particle concentration limitation of each area and the lowest system energy consumption as targets, the optimal air supply amount of each area is determined through a multi-agent distributed optimization algorithm, and accordingly, each FFU frequency conversion controller adjusts the corresponding FFU fan. Personnel states are perceived in real time through machine vision, the multi-area particulate matter concentration is predicted in combination with a mass balance equation and air flow exchange among different areas, dynamic cooperative regulation and control of the air supply amount of each FFU area are achieved through a multi-agent distributed optimization algorithm, and on the premise that it is ensured that the cleanliness of the whole area reaches the standard, the air supply amount of each FFU area is increased. The total air volume of the system is reduced to the maximum and the energy-saving target is achieved.

Description

A method, system, equipment and medium for multi-zone distributed intelligent control of cleanrooms Technical Field

[0001] This invention relates to the field of cleanroom environmental control and energy-saving technology. More specifically, it relates to the design of a machine vision-based multi-zone distributed intelligent control system and method for cleanrooms, which is particularly suitable for cleanrooms in industries such as electronics manufacturing, biomedicine, and precision instruments that have strict requirements for cleanliness and energy consumption. Background Technology

[0002] As the core environment for industrial production and scientific research, the cleanliness of cleanrooms directly determines product quality and experimental accuracy. Therefore, maintaining a specific cleanliness level within cleanrooms is a crucial prerequisite for ensuring the smooth operation of related activities. To achieve this goal, existing cleanrooms generally adopt a design mode of high air volume and uniform air supply throughout the entire area. By introducing a large amount of fresh air, the concentration of indoor particulate matter is diluted, thereby ensuring that the concentration of pollutants in the cleanroom meets the requirements.

[0003] Currently, cleanroom airflow control methods are mainly divided into two categories: one is the fixed airflow mode, which means that the air supply volume operates at a constant value according to the initial design value. Regardless of changes in the actual working conditions such as the number of people, their activity status, and production load, the air conditioning and ventilation system maintains full-load operation. The other is the shift mode, which is mainly used in scenarios such as pharmaceutical plants where there is a clear distinction between working and non-working periods. During non-working periods such as nighttime, the air supply volume is reduced to the minimum level that meets the basic cleanliness requirements, and during working periods, it is restored to the designed airflow volume to reduce energy consumption during non-working periods.

[0004] However, existing cleanroom airflow control methods suffer from significant energy waste and insufficient control precision, specifically in the following aspects: First, energy consumption is extremely high and waste is severe. The air conditioning and ventilation system of a cleanroom accounts for 60%-80% of its total energy consumption. Due to the need to reserve high redundancy during the design phase to cope with extreme working conditions, the actual airflow in the fixed airflow mode is often 1.5-2 times the minimum airflow required to meet cleanliness requirements. Even with a shift-based mode, energy consumption can only be reduced during non-working hours, while high airflow still needs to be maintained during working hours, resulting in significant overall energy waste.

[0005] Secondly, variable air volume (VAV) control technology has inherent limitations. In recent years, VAV operation methods have been gradually proposed to optimize energy consumption. However, traditional VAV control is mostly based on concentration sensors, which has two major drawbacks: First, the sensors are sparsely distributed, making it impossible to comprehensively and in real time reflect the differences in particulate matter concentration in multiple areas of the cleanroom, which can easily lead to blind spots in control. Second, the sensor signals have significant lag. When the particulate matter concentration exceeds the standard, the air volume is adjusted, which can no longer ensure that the cleanliness standard is met in time. Moreover, frequent air volume fluctuations will further increase energy consumption.

[0006] Secondly, the impact of personnel dynamics on cleanliness is not considered. Personnel are one of the main sources of particulate matter pollution in cleanrooms, and their number, activity intensity, and location directly determine the emission rate and diffusion range of particulate matter. Existing control systems have not established a correlation model between personnel status and particulate matter concentration, making it impossible to achieve precise airflow control based on changes in personnel dynamics. This generally results in an imbalance where excessive airflow is supplied to unoccupied areas and insufficient control is provided to densely populated areas, wasting energy and making it difficult to ensure uniformity of cleanliness.

[0007] Finally, the zoning control lacks coordination. Cleanrooms typically use multiple fan filter units (FFUs) for zoned air supply. Existing systems mostly control each FFU independently without considering the airflow coupling relationship between different areas. This can easily lead to problems such as excessive airflow in some areas resulting in energy waste, or insufficient airflow in some areas resulting in substandard cleanliness. It is impossible to achieve a synergy between achieving cleanliness standards and optimizing energy consumption across the entire area.

[0008] In summary, existing cleanroom airflow control technologies cannot simultaneously ensure cleanliness and optimize energy consumption. Their control precision, real-time performance, and coordination are insufficient, making it difficult to meet the high-precision, low-energy consumption requirements of modern industrial production and scientific research. Therefore, developing a cleanroom control solution capable of real-time sensing of multi-area environmental conditions and personnel dynamics, and achieving distributed, coordinated, and precise control, has become a pressing technical problem to be solved in this field. Summary of the Invention

[0009] To overcome the shortcomings of existing technologies and address the problems of high energy consumption, low accuracy, and poor coordination in current cleanroom airflow control systems, this invention proposes a machine vision-based multi-zone distributed intelligent control method, system, equipment, and medium for cleanrooms. This method uses machine vision to perceive the number, location, and activity type of personnel in real time. Combined with the mass balance equation and airflow exchange between different zones, it predicts the particulate matter concentration in multiple zones. A multi-agent distributed optimization algorithm is then used to achieve dynamic and coordinated control of the airflow in each FFU zone. This minimizes the total system airflow while ensuring that the cleanliness of all zones meets the standards, thus achieving energy savings.

[0010] The objective of this invention can be achieved through the following technical solutions.

[0011] A multi-zone distributed intelligent control method for cleanrooms includes the following steps: S1, real-time acquisition of cleanroom images to obtain personnel status information, including the number of personnel, real-time location information, and activity type. Different activity types correspond to different particulate matter emission source intensities. The particulate matter emission source intensities of personnel in each area of ​​the cleanroom are determined by combining the real-time location information of personnel; S2, based on the particulate matter emission source intensities of personnel in each area of ​​the cleanroom and the current air supply volume of each area, the particulate matter concentration of each area is calculated using a particulate matter concentration prediction model; S3, under the constraint that the particulate matter concentration in each area does not exceed the standard, with the goal of minimizing the particulate matter concentration limit in each area and the system energy consumption, the optimal air supply volume of each area and the total system air volume are determined by a multi-agent distributed optimization algorithm; S4, according to the optimal air supply volume of each area, the frequency converter controller of each FFU adjusts the fan speed or frequency of the corresponding FFU to achieve precise control of the air supply volume in each area of ​​the cleanroom.

[0012] Further, the specific process of step S1 is as follows: The camera acquires cleanroom images in real time. Image noise is removed using Gaussian filtering, and camera distortion is corrected through perspective transformation. Then, based on the YOLOv8 object detection algorithm, the number of people in the image is determined and their positions are outlined. Combining the camera's visual spatial calibration information, the relative position information of each person on the room plane is further determined using a monocular camera depth estimation method. Then, the YOLOv8-pose model is used to extract key points of the human skeleton from the real-time image with outlined personnel positions, thereby obtaining the motion time series of the key points. Based on this, a GRU-based classifier is established, and this classifier is used to analyze the motion time series of the key points of the human skeleton to obtain the type of personnel activity. According to experiments or literature, different activity types correspond to different particulate matter emission source intensities. Combined with the real-time position information of the personnel, the particulate matter emission source intensity of each area in the cleanroom is determined. Specifically, according to the principle of time-sharing and zone-based control, the cleanroom can be divided into several independently controllable areas according to the arrangement of FFUs in the air conditioning and ventilation system.

[0013] Furthermore, the particulate matter concentration prediction model in step S2 adopts a multi-region mass balance model, which is established as follows: a mass balance model is established with each region of the cleanroom as a control volume to form a multi-region mass balance model; wherein, the difference form of the mass balance model is shown in formula (1). (1), where i and j are the row number and column number of each region, respectively; Let (i,j) be the volume of the region. Let be the particulate matter concentration in the region (i,j); t is the time step; The air supply volume of the FFU in area (i,j); The concentration of particulate matter in the FFU supply air; The intensity of particulate matter emission sources from personnel within the (i,j) region; , , , These represent the particulate matter concentrations in the left, right, front, and rear regions of the current calculation region (i,j), respectively. , , , These represent the airflow exchange volume between the current calculation area and its adjacent areas to the left, right, front, and rear; where the FFU supply air particulate concentration is... It can be calculated using the following formula: (2), (3), where, This refers to the return air volume of the air conditioning and ventilation system. The fresh air volume of an air conditioning and ventilation system is the total fresh air volume. The supply air volume is the sum of the fresh air volume and the return air volume. This is also equal to the air supply volume of each area's FFU. sum; The concentration of particulate matter in fresh air; The overall filtration efficiency of the fresh air conditioning unit; The filtration efficiency of the FFU is given; the airflow exchange between adjacent areas can be estimated based on an approximate parallel airflow mixing process, as shown in the following formula: (4), where α is the mixing intensity coefficient and A is the interface area between two adjacent regions; Let be the average wind speed in region (i,j).

[0014] Furthermore, the expressions for the constraints and objective function in step S3 are as follows: (5), where, To achieve the target for particulate matter concentration control in cleanrooms, , Upper limit of particulate matter concentration; system energy consumption , For the air supply efficiency of FFU; These are the weighting coefficients related to particulate matter concentration. These are the weighting coefficients related to system energy consumption.

[0015] Furthermore, the specific process for determining the optimal air supply volume for each region in step S3 using a multi-agent distributed optimization algorithm is as follows: S31 Initialization: Initialize the state of the central coordinating agent and each FFU agent, and the current air supply volume of each FFU in each region. Upper limit of particulate matter concentration Intensity of particulate matter emission sources from people in each area The central coordinating agent is responsible for global coordination, with the goal of minimizing system energy consumption. Each FFU agent corresponds one-to-one with each FFU in the cleanroom. The FFU agent reduces the airflow of each FFU to ensure that the particulate matter concentration in its area does not exceed the standard. At the same time, it receives global instructions from the central coordinating agent and feeds back the local optimization results to the central coordinating agent. S32 Local optimization: The particulate matter concentration in each area is solved using the above formula (5). Determine whether it meets the requirements. FFU supply air volume in each area under constraints Minimum value; S33 Global Coordination: Centrally coordinates and intelligently collects the air volume of FFUs in each area. The amount of airflow exchanged between regions , , , The update is transmitted to each FFU agent, which then performs further local optimization. Through multiple iterations, the optimal air delivery rate for each FFU area is obtained. S34 command issued: Each FFU agent will optimize airflow. The command is sent to each FFU frequency converter controller.

[0016] The objective of this invention can also be achieved through the following technical solutions.

[0017] A multi-zone distributed intelligent control system for cleanrooms includes: a machine vision perception and recognition module, which acquires cleanroom images in real time and obtains personnel status information, including the number of personnel, their real-time location information, and their activity types. Different personnel activity types correspond to different particulate matter emission source intensities. The system determines the particulate matter emission source intensity of each area of ​​the cleanroom based on the personnel's real-time location information; a particulate matter concentration prediction module, which calculates the particulate matter concentration of each area based on the particulate matter emission source intensity of each area and the current air supply volume of each area using a particulate matter concentration prediction model; and a multi-agent distributed optimization decision module, which, under the constraint that the particulate matter concentration of each area does not exceed the standard, aims to minimize the particulate matter concentration limit of each area and the system energy consumption. The module determines the optimal air supply volume of each area and the total system air volume through a multi-agent distributed optimization algorithm, and sends the optimal air supply volume of each area to the corresponding FFU frequency converter. Each FFU frequency converter adjusts the fan speed or frequency of the corresponding FFU to achieve precise control of the air supply volume of each area of ​​the cleanroom.

[0018] Furthermore, the machine vision perception and recognition module includes a camera, an image preprocessing unit, a personnel position tracking module, and a personnel behavior recognition submodule. The camera is used to acquire cleanroom images in real time and transmit them to the image preprocessing unit. The image preprocessing unit first uses Gaussian filtering to remove noise from the image, and then corrects camera distortion through perspective transformation. The preprocessed image is then transmitted to the personnel position tracking module. The personnel position tracking module determines the number of personnel in the image based on the YOLOv8 target detection algorithm, outlines the real-time positions of the personnel in the image, and further determines the relative position information of each person on the room plane using a monocular camera depth estimation method, combined with the camera's visual spatial calibration information, to achieve real-time tracking of personnel positions. The personnel behavior recognition submodule... YOLOv8-pose was used to extract key points of the human skeleton from real-time images with the personnel positions outlined, thereby obtaining the motion time series of the key points. Based on this, a GRU-based classifier was built, and this classifier was used to analyze the motion time series of the key points of the human skeleton to obtain the type of personnel activity. According to the different activity types obtained from experiments or literature, different particulate matter emission source intensities correspond to different activity types. Combined with the real-time location information of personnel, the particulate matter emission source intensity of personnel in each area of ​​the cleanroom was determined. Finally, the number of personnel, real-time location information of personnel, and particulate matter emission source intensity of personnel in each area were transmitted to the particulate matter concentration prediction module in real time. In particular, according to the time-sharing and zone-based control principle, the cleanroom was divided into several independently controllable areas according to the arrangement of FFUs in the air conditioning and ventilation system.

[0019] Furthermore, the multi-intelligent distributed optimization decision-making module consists of a central coordinating agent and several FFU agents. The central coordinating agent is responsible for global coordination, with the goal of minimizing system energy consumption. Each FFU agent corresponds one-to-one with each FFU in the cleanroom. The FFU agent reduces the airflow of each FFU while ensuring that the particulate matter concentration in its area does not exceed the standard. At the same time, it receives global instructions from the central coordinating agent and feeds back the local optimization results to the central coordinating agent.

[0020] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described cleanroom multi-zone distributed intelligent control method.

[0021] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described cleanroom multi-zone distributed intelligent control method.

[0022] Compared with the prior art, the beneficial effects of the technical solution of the present invention are: (1) Significantly reduce energy consumption and achieve energy-saving and efficient operation: The present invention breaks through the redundant design limitations of the existing fixed air volume mode. It uses machine vision to perceive the dynamics of personnel (number, location, activity type) and the environmental status of multiple areas in the clean room in real time. Combined with the particulate matter concentration correlation model, it dynamically matches the minimum air supply demand, avoiding the problem of excessive air supply in unmanned areas and high air volume waste in sparsely populated areas. At the same time, through the distributed collaborative control of multiple fan filter units (FFU), it avoids the energy consumption redundancy caused by local air volume superposition. Compared with the existing fixed air volume mode, it can reduce the energy consumption of the ventilation and air conditioning system by 30%-50%. Even compared with the duty mode, it can further reduce the energy consumption during working hours by more than 20%, significantly improving the economic efficiency of clean room operation.

[0023] (2) Improve the accuracy and real-time performance of regulation and ensure stable cleanliness standards: This invention abandons the traditional regulation method based on sparse concentration sensors and adopts machine vision technology to achieve full-area coverage monitoring of the cleanroom without dead angles. It can capture the differences and dynamic trends of particulate matter concentration in multiple areas in real time, and there is no problem of sensor signal lag. By establishing a correlation model between personnel status and particulate matter emission and diffusion, it can predict the risk of cleanliness changes in advance and accurately adjust the air volume of the corresponding area, so that the cleanliness of each area is always stably controlled within the set threshold. This effectively avoids the lag defect of regulation after the concentration exceeds the standard, significantly improves the reliability of the cleanroom environment, and ensures the quality of precision production and scientific experiments.

[0024] (3) Adapting to dynamic working conditions and improving system flexibility and applicability: The present invention can respond in real time to the dynamic changes in the number of people, activity intensity, and regional distribution in the clean room, as well as the differences in cleanliness requirements caused by production load fluctuations. It achieves refined and differentiated allocation of air volume through distributed control strategy. Compared with the existing fixed control mode or single duty mode, it is suitable for a variety of complex working conditions. Whether it is a production scenario with frequent personnel flow such as semiconductor manufacturing and precision electronic processing, or a scientific research scenario with variable personnel activity intensity such as biological experiments, it can stably exert the control effect, greatly improving the applicability and flexible adjustment capability of the system.

[0025] (4) Achieving multi-regional coordinated control to ensure uniformity of cleanliness: This invention breaks through the limitations of independent control of each FFU in the existing system. By constructing an inter-regional airflow coupling model through a distributed control architecture, it coordinates the air supply parameters of each zone, avoiding the problem of insufficient cleanliness caused by local air volume or energy waste caused by local air volume superposition. Combining distributed optimization with multi-agent systems not only ensures the independence of control in each region, but also achieves the global energy-saving goal. Compared with traditional centralized optimization algorithms, the decision-making efficiency is improved and the energy-saving effect is more significant. It can ensure that the air velocity and particulate matter concentration in each region of the clean room are uniform, effectively eliminating cleanliness blind spots and local pollution accumulation, comprehensively improving the uniformity of the overall environmental quality of the clean room, and providing a stable environmental guarantee for high-precision production and experimentation in the entire region.

[0026] (5) Simplify system maintenance and reduce operation and maintenance costs: Compared with traditional multi-point concentration sensor arrays, the machine vision monitoring scheme adopted in this invention has the advantages of easy installation, low failure rate and low maintenance cost. It does not require frequent calibration and replacement of sparsely distributed sensors. At the same time, the intelligent control logic of the system can automatically adapt to changes in working conditions, reduce the frequency of manual intervention, reduce the workload and operation and maintenance costs of maintenance personnel, and further improve the level of intelligence of cleanroom management.

[0027] In summary, this invention effectively solves the core pain points of existing cleanroom control technologies, such as high energy consumption, low accuracy, poor real-time performance, and insufficient adaptability, through the innovative combination of machine vision perception and distributed collaborative control. It achieves a synergistic unity between cleanliness assurance and energy consumption optimization, demonstrating significant technological advancement and practical application value, and can be widely applied to various high-precision cleanroom scenarios. Attached Figure Description

[0028] Figure 1 is a flowchart of the multi-zone distributed intelligent control method for cleanrooms according to the present invention.

[0029] Figure 2 is a schematic diagram of the cleanroom air conditioning and ventilation system layout in this invention.

[0030] Figure 3 is a top view of the cleanroom partitioning of the present invention.

[0031] Figure 4 is a schematic diagram of the cleanroom multi-zone distributed intelligent control system of the present invention.

[0032] Figure 5 is an example diagram of the multi-agent distributed control architecture in this invention.

[0033] Figure 6 is a flowchart of the workflow of the cleanroom multi-zone distributed intelligent control system of the present invention. Detailed Implementation

[0034] The present invention will now be further described with reference to the accompanying drawings.

[0035] This invention proposes a machine vision-based multi-zone distributed intelligent control method for cleanrooms, as shown in Figure 1. The method includes the following steps: S1 Visual Perception and Recognition: Real-time acquisition of cleanroom images to obtain personnel status information, including the number of personnel, their real-time location, and their activity type. Different activity types correspond to different particulate matter emission source intensities. The particulate matter emission source intensity in each area of ​​the cleanroom is determined by combining the real-time location information. S2 Particulate Matter Concentration Prediction: Based on the particulate matter emission source intensity in each area of ​​the cleanroom and the current airflow in each area, a particulate matter concentration prediction model is used to calculate the particulate matter concentration in each area. S3 Multi-Agent Distributed Optimization: Under the constraint that the particulate matter concentration in each area does not exceed the limit, and with the goal of minimizing particulate matter concentration and system energy consumption, a multi-agent distributed optimization algorithm is used to determine the optimal airflow in each area and the total system airflow. S4 FFU Variable Frequency Controller Adjustment: Based on the optimal airflow in each area, the variable frequency controller of each FFU adjusts the fan speed or frequency of the corresponding FFU to achieve precise control of the airflow in each area of ​​the cleanroom.

[0036] In one possible implementation, preferably, step S1 involves the following steps: A camera acquires real-time images of the cleanroom; image noise is removed using Gaussian filtering; camera distortion is corrected using perspective transformation; then, the number of people in the image is determined and their positions are outlined using the YOLOv8 object detection algorithm; combined with the camera's visual spatial calibration information, the relative position information of each person on the room plane is further determined using a monocular camera depth estimation method; then, the YOLOv8-pose model is used to extract key points of the human skeleton from the real-time images with outlined personnel positions, thereby obtaining a time series of motion of the key points of the human skeleton; based on this, a GRU-based classifier is established; this classifier is used to analyze the time series of motion of the key points of the human skeleton to obtain the type of personnel activity; according to experiments or literature, different activity types correspond to different particulate matter emission source intensities; and combined with the real-time position information of personnel, the particulate matter emission source intensity of each area in the cleanroom is determined. Furthermore, according to the principle of time-sharing and zone-based control, the cleanroom is divided into several independently controllable areas according to the arrangement of FFUs in the air conditioning and ventilation system. A schematic diagram of the air conditioning and ventilation system layout is shown in Figure 2, and a top-down plan view of the zones is shown in Figure 3.

[0037] In one possible implementation, preferably, the particulate matter concentration prediction model in step S2 adopts a multi-region mass balance model. The establishment process is as follows: a mass balance model is established with each region of the cleanroom as the control body, and the resulting set of equations is the multi-region mass balance model, which accurately predicts the real-time particulate matter concentration in each region of the cleanroom and provides data support for control decisions.

[0038] For cleanrooms divided into multiple independent control zones based on the number of FFUs, the dynamic change of particulate matter concentration in each zone follows the following equation: (1), where i and j are the row number and column number of each region, respectively; Let (i,j) be the volume of the region. Let be the particulate matter concentration in the region (i,j); t is the time step; The air supply volume of the FFU in area (i,j) is equal to the air outlet volume through the floor in that area; The concentration of particulate matter in the FFU supply air; The intensity of particulate matter emission sources from personnel within the (i,j) region; Let be the particulate matter settling coefficient in region (i,j), which can usually be ignored; , , , These represent the particulate matter concentrations in the left, right, front, and rear regions of the current calculation region (i,j), respectively. , , , These represent the airflow exchange between the current calculation area and its left, right, front, and rear adjacent areas, respectively; for areas located at the edges or corners, this value is not applicable and can be set to 0.

[0039] Among them, the particulate matter concentration in the FFU air supply It can be calculated using the following formula: (2), (3), where, This refers to the return air volume of the air conditioning and ventilation system. The fresh air volume of an air conditioning and ventilation system is the total fresh air volume. The supply air volume is the sum of the fresh air volume and the return air volume. This is also equal to the air supply volume of each area's FFU. sum; The concentration of particulate matter in the fresh air is assumed to be uniformly mixed in all areas. Therefore, the concentration of particulate matter in the return air of a region is related to the concentration of particulate matter in that region. Consistent; The overall filtration efficiency of the fresh air conditioning unit; This refers to the filtration efficiency of the FFU.

[0040] The amount of airflow exchange between adjacent areas can be estimated based on an approximately parallel airflow mixing process, as shown in the following formula: (4), where α is the mixing intensity coefficient, which reflects the efficiency of airflow exchange and can usually be estimated through experience or CFD simulation; A is the interface area between two adjacent regions. Let be the average wind speed in region (i,j).

[0041] To simplify the calculation, particulate matter sedimentation can be ignored. The mass balance model in formula (1) can be modified into a difference form, as shown in formula (5). (5), where, t is the time step.

[0042] This invention utilizes machine vision to dynamically identify and classify human behavior, dynamically acquiring the type of human activity and the intensity of particulate matter emission sources. Based on a multi-regional mass balance model, it introduces inter-regional airflow exchange to correct for the influence of particulate matter diffusion in adjacent regions, thereby improving the accuracy of particulate matter concentration prediction. This invention considers only human activity as the primary pollution source; other pollution sources can be added to the multi-regional mass balance model in a similar manner.

[0043] In one possible implementation, preferably, the expressions for the constraints and the objective function in step S3 are as follows: (6) Considering the strict requirements for particulate matter concentration in cleanrooms, the target for particulate matter concentration control in cleanrooms is as follows: It can be represented as: , This represents the upper limit for particulate matter concentration. System energy consumption. , This refers to the air supply efficiency of the FFU (Fan Filter Unit). These are the weighting coefficients related to particulate matter concentration. The importance of the two objectives can be determined by the weighting coefficients related to system energy consumption. and Adjustment.

[0044] In one possible implementation, preferably, the specific process of determining the optimal air supply volume for each region in step S3 using a multi-agent distributed optimization algorithm is as follows: S31 Initialization: Initialize the state of the central coordinating agent and each FFU agent, and the current air supply volume of each FFU in each region. Upper limit of particulate matter concentration Intensity of particulate matter emission sources from people in each area .

[0045] The central coordinating agent is responsible for overall coordination, with the goal of minimizing system energy consumption. Each FFU agent corresponds one-to-one with each FFU in the cleanroom. The FFU agent reduces the airflow of each FFU while ensuring that the particulate matter concentration in its area does not exceed the standard. At the same time, it receives global instructions from the central coordinating agent and feeds back the local optimization results to the central coordinating agent.

[0046] S32 Local Optimization: Solve for the particulate matter concentration in each region using the above formula (6). Determine whether it meets the requirements. FFU supply air volume in each area under constraints The minimum value.

[0047] S33 Global Coordination: Centrally coordinates and intelligently collects the air volume of FFUs in each area. The amount of airflow exchanged between regions , , , The update is transmitted to each FFU agent, which then performs further local optimization. Through multiple iterations, the optimal air delivery rate for each FFU area is obtained. .

[0048] S34 command issued: Each FFU agent will optimize airflow. The command is sent to each FFU frequency converter controller.

[0049] Based on the principles of the above-mentioned multi-zone distributed intelligent control method for cleanrooms, this invention also proposes a multi-zone distributed intelligent control system for cleanrooms, as shown in Figure 4. It mainly includes a machine vision perception and recognition module, a particulate matter concentration prediction module, and a multi-intelligent distributed optimization decision-making module. The modules work together to achieve intelligent control of air volume in multiple zones of the cleanroom.

[0050] The core function of the machine vision perception and recognition module is to acquire cleanroom images in real time and comprehensively, obtaining personnel status information within the cleanroom, including the number of personnel, their real-time location information, and their activity types. Different personnel activity types correspond to different particulate matter emission source intensities. By combining the real-time location information of personnel, the particulate matter emission source intensity in each area of ​​the cleanroom is determined, providing core input parameters for subsequent concentration prediction in different areas of the cleanroom. Specifically, based on the principle of time-sharing and zone-based control, the cleanroom is divided into several independently controllable zones according to the arrangement of FFUs in the air conditioning and ventilation system.

[0051] The machine vision perception and recognition module includes a camera, an image preprocessing unit, a personnel position tracking module, and a personnel behavior recognition submodule. The camera is used to acquire cleanroom images in real time and transmit them to the image preprocessing unit. High-definition industrial cameras are installed on the cleanroom ceiling and in corners to ensure that each FFU (Fan Filter Unit) air supply area is within the camera's visual monitoring range, adhering to the principle of no blind spots. The image preprocessing unit first uses Gaussian filtering to remove noise from the image, then corrects camera distortion through perspective transformation to improve image quality. The preprocessed image is then transmitted to the personnel position tracking module. The personnel position tracking module uses the YOLOv8 object detection algorithm to determine the number of personnel in the image and outlines their real-time positions. Combining this with the camera's visual spatial calibration information, it further uses a monocular camera depth estimation method to determine the relative position information of each person on the room plane, achieving real-time tracking of personnel positions. The personnel behavior recognition submodule uses YOLOv8-pose to extract key points of the human skeleton from real-time images with the personnel positions outlined, thereby obtaining the motion time series of the key points. Based on this, a GRU-based classifier is built, and this classifier analyzes the motion time series of the key points to determine the type of personnel activity (e.g., sitting / standing, walking, knee bending, arm extension, etc.). According to experimental or literature findings, different activity types correspond to different particulate matter emission source intensities. Combined with real-time personnel location information, the particulate matter emission source intensity in each area of ​​the cleanroom is determined. Finally, the number of personnel, their real-time location information, and the particulate matter emission source intensity in each area are transmitted in real-time to the particulate matter concentration prediction module. The machine vision perception and recognition module, through multi-algorithm fusion, solves the problems of limited monitoring range and lag in monitoring and control inherent in traditional single particulate matter sensors.

[0052] Cleanrooms have extremely high requirements for the concentration of particulate matter in their rooms. The main task of the cleanroom air conditioning and ventilation system is to reduce the particulate matter concentration to below the standard required value. Therefore, predicting the indoor particulate matter concentration under different operating conditions is an important support and criterion for cleanroom control decisions. This invention provides the intensity of particulate matter emission sources from personnel in different areas based on machine vision-perceived personnel status information. According to the time-sharing and zoning control principle, the cleanroom is divided into several independently controllable areas according to the arrangement of FFUs in the air conditioning and ventilation system. For the particulate matter concentration prediction module in the zoning situation, a multi-area mass balance model is first established as the particulate matter concentration prediction model, as shown in the process of formulas (1) to (5) above. Then, based on the intensity of particulate matter emission sources from personnel in each area of ​​the cleanroom and the current air supply volume in each area, the real-time particulate matter concentration in each area of ​​the cleanroom is accurately calculated using the particulate matter concentration prediction model, providing data support for control decisions.

[0053] The multi-agent distributed optimization decision module is the core of the system's decision-making process. Its core function is to determine the optimal air supply volume for each region and the total system air volume under the constraint that the particulate matter concentration in each region does not exceed the standard. The objective function and constraint expressions are shown in the above formula (6). The optimal air supply volume for each region and the total system air volume are determined by the multi-agent distributed optimization algorithm (see the above optimization steps S31~S34): Each FFU agent solves for the air supply volume for its respective region. The minimum value is determined by the central coordinating agent, which coordinates globally based on regional coupling relationships to reduce system energy consumption. Then, each FFU agent further optimizes the airflow of the FFUs in each region. Through multiple iterations of optimization and coordination, the goal of minimizing system energy consumption was achieved while ensuring that particulate matter concentrations in each area do not exceed limits. During the iteration process, the FFU (Fan Filter Unit) airflow in each area was continuously adjusted. The data is then passed to the particulate matter concentration prediction model to calculate the particulate matter concentration in each zone under different airflow conditions. Meanwhile, particulate matter concentration prediction parameters are used in the optimization decision-making process of each agent. Finally, the optimal air supply volume for each area is sent to the corresponding FFU frequency converter, and each FFU frequency converter adjusts the fan speed or frequency of the corresponding FFU to achieve precise control of the air supply volume in each area of ​​the cleanroom and achieve the goal of energy saving.

[0054] As shown in Figure 5, the multi-intelligent distributed optimization decision-making module consists of a central coordinating agent and several FFU agents. The central coordinating agent is responsible for global coordination, with the goal of minimizing system energy consumption. Each FFU agent corresponds one-to-one with each FFU in the cleanroom. The FFU agent reduces the airflow of each FFU while ensuring that the particulate matter concentration in its area does not exceed the standard. Simultaneously, it receives global instructions from the central coordinating agent and feeds back the local optimization results to the central coordinating agent.

[0055] As shown in Figure 6, after the cleanroom multi-zone distributed intelligent control system of the present invention is started, it first initializes the working parameters of each module: setting the volume of each area of ​​the cleanroom. Fresh air volume Fresh air particulate matter concentration Basic parameters; setting the upper limit of particulate matter concentration. The camera position coordinates are calibrated, and areas within the camera's detection range are divided to correspond with the areas where the FFUs are located, ensuring a complete match between the visual monitoring range and the FFU air supply area. Then, the machine vision perception and recognition module performs real-time perception of personnel status to determine the intensity of particulate matter emission sources in each area. Subsequently, the particulate matter concentration prediction module accurately predicts the real-time particulate matter concentration in each area of ​​the cleanroom based on the intensity of particulate matter emission sources from personnel and the current air supply volume in each area. This prediction is then transmitted to the multi-agent distributed optimization decision-making module. This module, aiming to meet the particulate matter concentration limits in each area and minimize system energy consumption, uses a multi-agent distributed optimization algorithm to solve for the optimal air supply volume of the FFUs in each area. Ultimately, the optimal air supply volume for each area will be determined. The data is sent to the corresponding FFU frequency converters to adjust the fan speed or frequency of the corresponding FFU, thereby achieving precise control of the air supply volume in each area of ​​the cleanroom.

[0056] Based on the above principles, the present invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-mentioned cleanroom multi-zone distributed intelligent control method.

[0057] Based on the above principles, the present invention also proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-mentioned cleanroom multi-zone distributed intelligent control method.

[0058] Although the functions and working processes of the present invention have been described above in conjunction with the accompanying drawings, the present invention is not limited to the specific functions and working processes described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these are within the protection scope of the present invention.

Claims

1. A method for multi-zone distributed intelligent control of a cleanroom, characterized in that, Includes the following steps: S1. Real-time acquisition of cleanroom images to obtain personnel status information, including the number of personnel, their real-time location information, and their activity type. Different personnel activity types correspond to different particulate matter emission source intensities. Combining the real-time location information of personnel, the particulate matter emission source intensity of each area of ​​the cleanroom is determined. S2. Based on the particulate matter emission source intensity of each area of ​​the cleanroom and the current air supply volume of each area, the particulate matter concentration of each area is calculated using a particulate matter concentration prediction model. S3. Under the constraint that the particulate matter concentration of each area does not exceed the standard, with the goal of minimizing the particulate matter concentration limit of each area and the system energy consumption, the optimal air supply volume of each area and the total system air volume are determined through a multi-agent distributed optimization algorithm. S4, based on the optimal air supply volume for each area, each FFU frequency converter adjusts the fan speed or frequency of the corresponding FFU to achieve precise control of the air supply volume in each area of ​​the cleanroom.

2. The cleanroom multi-zone distributed intelligent control method according to claim 1, characterized in that, The specific process of step S1 is as follows: The camera acquires cleanroom images in real time, removes image noise through Gaussian filtering, corrects camera distortion through perspective transformation, then determines the number of people in the image and outlines their positions based on the YOLOv8 target detection algorithm, and further determines the relative position information of each person on the room plane by combining the camera's visual spatial calibration information with the monocular camera depth estimation method. Then, the YOLOv8-pose model is used to extract human skeleton key points from the real-time image with outlined personnel positions, thereby obtaining the motion time series of human skeleton key points. On this basis, a GRU-based classifier is established, and this classifier is used to analyze the motion time series of human skeleton key points to obtain the type of personnel activity. According to the different activity types obtained from experiments or literature, different particulate matter emission source intensities correspond to different activity types. Combined with the real-time position information of personnel, the particulate matter emission source intensity of personnel in each area of ​​the cleanroom is determined. Among them, according to the time-sharing and zone-based control principle, the cleanroom is divided into several independently controllable areas according to the arrangement of FFUs in the air conditioning and ventilation system.

3. The cleanroom multi-zone distributed intelligent control method according to claim 1, characterized in that, The particulate matter concentration prediction model in step S2 adopts a multi-region mass balance model. The establishment process is as follows: a mass balance model is established with each region of the cleanroom as a control volume to form a multi-region mass balance model; wherein, the difference form of the mass balance model is shown in formula (1). (1), where i and j are the row number and column number of each region, respectively; Let (i,j) be the volume of the region. Let be the particulate matter concentration in the region (i,j); t is the time step; The air supply volume of the FFU in area (i,j); The concentration of particulate matter in the FFU supply air; The intensity of particulate matter emission sources from personnel within the (i,j) region; 、 、 、 These represent the particulate matter concentrations in the left, right, front, and rear regions of the current calculation region (i,j), respectively. 、 、 、 These represent the airflow exchange volume between the current calculation area and its adjacent areas to the left, right, front, and rear; where the FFU supply air particulate concentration is... It can be calculated using the following formula: （2）， (3), where, This refers to the return air volume of the air conditioning and ventilation system. The fresh air volume of an air conditioning and ventilation system is the total fresh air volume. The supply air volume is the sum of the fresh air volume and the return air volume. This is also equal to the air supply volume of each area's FFU. Yukiwa; The concentration of particulate matter in fresh air; The overall filtration efficiency of the fresh air conditioning unit; The filtration efficiency of the FFU is given; the airflow exchange between adjacent areas can be estimated based on an approximate parallel airflow mixing process, as shown in the following formula: (4), where α is the mixing intensity coefficient and A is the interface area between two adjacent regions; Let be the average wind speed in region (i,j).

4. The cleanroom multi-zone distributed intelligent control method according to claim 1, characterized in that, The expressions for the constraints and objective function in step S3 are as follows: (5), where, To achieve the target for particulate matter concentration control in cleanrooms, ， Upper limit of particulate matter concentration; system energy consumption ， For the air supply efficiency of FFU; These are the weighting coefficients related to particulate matter concentration. These are the weighting coefficients related to system energy consumption.

5. The cleanroom multi-zone distributed intelligent control method according to claim 1, characterized in that, The specific process for determining the optimal air supply volume for each region in step S3 using a multi-agent distributed optimization algorithm is as follows: S31 Initialization: Initialize the state of the central coordinating agent and each FFU agent, and the current air supply volume of each FFU in each region. Upper limit of particulate matter concentration Intensity of particulate matter emission sources from people in each area The central coordinating agent is responsible for global coordination, with the goal of minimizing system energy consumption. Each FFU agent corresponds one-to-one with each FFU in the cleanroom. The FFU agent reduces the airflow of each FFU to ensure that the particulate matter concentration in its area does not exceed the standard. At the same time, it receives global instructions from the central coordinating agent and feeds back the local optimization results to the central coordinating agent. S32 Local optimization: The particulate matter concentration in each area is solved using the above formula (5). Determine whether it meets the requirements. FFU supply air volume in each area under constraints Minimum value; S33 Global Coordination: Centrally coordinates and intelligently collects the air volume of FFUs in each area. The amount of airflow exchanged between regions 、 、 、 The update is transmitted to each FFU agent, which then performs further local optimization. Through multiple iterations, the optimal air delivery rate for each FFU area is obtained. S34 command issued: Each FFU agent will optimize airflow. The command is sent to each FFU frequency converter controller.

6. A cleanroom multi-zone distributed intelligent control system based on the cleanroom multi-zone distributed intelligent control method according to any one of claims 1 to 5, characterized in that, include: The machine vision perception and recognition module acquires real-time images of the cleanroom and obtains personnel status information, including the number of personnel, their real-time location information, and their activity types. Different personnel activity types correspond to different particulate matter emission source intensities. Combining the real-time location information of personnel, the module determines the particulate matter emission source intensity of each area of ​​the cleanroom. The particulate matter concentration prediction module calculates the particulate matter concentration of each area based on the particulate matter emission source intensity of each area and the current air supply volume of each area using a particulate matter concentration prediction model. The multi-agent distributed optimization decision module, under the constraint that the particulate matter concentration of each area does not exceed the standard, aims to minimize the particulate matter concentration limit of each area and the system energy consumption. It determines the optimal air supply volume of each area and the total system air volume through a multi-agent distributed optimization algorithm. The optimal air supply volume of each area is then distributed to the corresponding FFU frequency converter. Each FFU frequency converter adjusts the fan speed or frequency of the corresponding FFU to achieve precise control of the air supply volume of each area of ​​the cleanroom.

7. The cleanroom multi-zone distributed intelligent control system according to claim 6, characterized in that, The machine vision perception and recognition module includes a camera, an image preprocessing unit, a personnel location tracking module, and a personnel behavior recognition submodule. The camera is used to acquire cleanroom images in real time and transmit them to the image preprocessing unit. The image preprocessing unit first uses Gaussian filtering to remove noise from the image, then corrects camera distortion through perspective transformation. The preprocessed image is then transmitted to the personnel location tracking module. The personnel location tracking module determines the number of personnel in the image based on the YOLOv8 target detection algorithm. The system outlines the real-time positions of personnel in the image and, combined with the camera's visual spatial calibration information, further determines the relative positions of each person on the room plane using a monocular camera depth estimation method, achieving real-time tracking of personnel positions. The personnel behavior recognition submodule uses YOLOv8-pose to extract key points of the human skeleton from the real-time image with the outlined personnel positions, thereby obtaining the motion time series of the key points of the human skeleton. Based on this, a GRU-based classifier is established, and this classifier is used to analyze the motion time series of the key points of the human skeleton to obtain the type of personnel activity. According to the different activity types obtained from experiments or literature, different particulate matter emission source intensities correspond to different activity types. Combined with the real-time personnel position information, the particulate matter emission source intensity of personnel in each area of ​​the cleanroom is determined. Finally, the number of personnel, real-time personnel position information, and particulate matter emission source intensity of personnel in each area are transmitted to the particulate matter concentration prediction module in real time. Among them, according to the time-sharing and zone-based control principle, the cleanroom is divided into several independently controllable areas according to the arrangement of FFUs in the air conditioning and ventilation system.

8. The cleanroom multi-zone distributed intelligent control system according to claim 6, characterized in that, The multi-intelligent distributed optimization decision-making module consists of a central coordinating agent and several FFU agents. The central coordinating agent is responsible for global coordination, with the goal of minimizing system energy consumption. Each FFU agent corresponds one-to-one with each FFU in the cleanroom. The FFU agent reduces the airflow of each FFU while ensuring that the particulate matter concentration in its area does not exceed the standard. At the same time, it receives global instructions from the central coordinating agent and feeds back the local optimization results to the central coordinating agent.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the cleanroom multi-zone distributed intelligent control method as described in any one of claims 1 to 5.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the cleanroom multi-zone distributed intelligent control method as described in any one of claims 1 to 5.

Images