Clean room control method and system adopting intelligent algorithm

Abstract

The invention discloses a clean room control method and system adopting an intelligent algorithm, and belongs to the technical field of clean air conditioning systems. The method comprises the steps that real-time environment parameters and equipment operation parameters in a clean room are collected; inputting the parameters into a pre-trained clean room neural network model; the cleanliness and the pressure difference are used as constraint conditions, the system energy consumption is used as an optimization target, optimization calculation is carried out through an intelligent algorithm, and optimal control parameters are obtained; and dynamically adjusting the running state of the clean room air system according to the optimal control parameters. According to the method, the neural network driven by the data and the intelligent optimization algorithm are fused, rapid stabilization and energy consumption collaborative optimization of the clean room under dynamic interference are achieved, and system operation energy consumption is remarkably reduced on the premise that it is guaranteed that environmental parameters reach the standard.

Description

Technical Field

[0001] This invention belongs to the field of cleanroom air conditioning system control technology, specifically relating to a method for dynamic adjustment of the cleanroom environment, and more specifically, to a cleanroom control method and system that uses intelligent algorithms and neural network models for collaborative optimization. Background Technology

[0002] For cleanrooms, especially those in the biopharmaceutical and electronics industries, the stable control of their internal environmental parameters (such as cleanliness, pressure differential, temperature, and humidity) is crucial. These environmental parameters directly affect product quality and production safety. As the core equipment for maintaining these parameters, the cleanroom air conditioning system accounts for a significant portion of the factory's total energy consumption, with the ventilation system being particularly energy-intensive.

[0003] Traditional cleanroom ventilation system control methods mainly rely on operating strategies based on fixed design airflow or simple feedback control (such as PID control). While these methods can maintain basic stability of environmental parameters under static or quasi-static conditions, they have significant shortcomings: First, they cannot effectively respond to dynamic disturbances such as personnel movement and equipment start-up and shutdown, resulting in large fluctuations in environmental parameters and long recovery times; second, because the rooms in a cleanroom are interconnected through air ducts, local adjustments (such as individual room air valves) can easily trigger cascading fluctuations in pressure differential and cleanliness in other rooms, increasing control complexity; finally, to conservatively ensure that cleanliness and pressure differential requirements are still met under the most unfavorable conditions, traditional methods typically employ a "full airflow" or constant high airflow operating mode, leading to significant energy waste.

[0004] In recent years, with the popularization of the Internet of Things and data acquisition technologies, some studies have attempted to introduce data analysis to optimize control. For example, historical operating data is collected to fit simple empirical models or set more refined fixed operating curves. However, these methods generally lack in-depth modeling of the complex dynamic mechanisms of cleanroom ventilation systems, resulting in limited control accuracy, robustness, and generalization ability, making it difficult to achieve truly global, real-time, and adaptive optimization.

[0005] Therefore, there is an urgent need in this field for an advanced control method that can deeply integrate system physical characteristics and real-time operational data, rapidly suppress dynamic disturbances, accurately maintain environmental parameters in multiple rooms, and minimize system energy consumption. This invention aims to solve the above problems by providing a dynamic optimization control scheme based on intelligent algorithms and a cleanroom neural network model. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the shortcomings of existing cleanroom control methods, such as slow response, high energy consumption and poor global coordination, and to provide a cleanroom control method and system that can be dynamically optimized in real time and take into account both environmental protection and energy saving.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0008] A cleanroom control method employing intelligent algorithms, characterized by comprising the following steps:

[0009] S1: Collect real-time environmental parameters and equipment operating parameters within the cleanroom;

[0010] S2: Input the environmental parameters and equipment operating parameters into a pre-trained cleanroom neural network model;

[0011] S3: Using cleanliness and pressure difference as constraints and system energy consumption as the optimization objective, the optimal control parameters with the lowest system energy consumption are obtained through intelligent algorithm optimization calculation.

[0012] S4: Dynamically adjust the operating status of the cleanroom ventilation system according to the optimal control parameters.

[0013] Furthermore, the intelligent algorithm used in step S3 is an optimization algorithm. The optimization algorithm inputs the combination of control parameters into the cleanroom neural network model for forward simulation, predicts the system performance indicators, and performs iterative optimization based on a fitness function that includes system energy consumption, pressure difference constraints, and cleanliness constraints.

[0014] More specifically, the workflow includes: taking a combination of various control parameters, including fan frequency and the opening degree of each air valve, as candidate solutions ("particles"), inputting them into the cleanroom neural network model for forward simulation, and quickly predicting the system's total power consumption, pressure difference in each room, and particle concentration under the corresponding scheme; then, evaluating the merits of the candidate solutions based on a fitness function that integrates energy consumption, pressure difference deviation, and cleanliness exceedance; and finally outputting the control parameter combination with the highest fitness, i.e., the system energy consumption is approximately minimized under the premise of strictly satisfying all environmental constraints.

[0015] Furthermore, the cleanroom neural network model is constructed using a data-driven approach and possesses online self-learning capabilities; when the deviation between the predicted value and the measured value exceeds a preset threshold or reaches a preset period, the system adaptively corrects the model parameters based on incremental data.

[0016] Furthermore, the cleanroom neural network model is a deep learning model based on the fusion architecture of KAN network and multi-task learning. Its inputs include cleanroom physical parameters, equipment parameters, real-time operating parameters and environmental parameters, and its outputs include total system air volume, fan frequency and the opening degree of air valves in each room.

[0017] More specifically, the cleanroom neural network model is a data-driven end-to-end deep learning model. Preferably, the model adopts a KAN network and multi-task learning fusion architecture, including an input module, a feature extraction module, and a multi-task output module. The input module receives heterogeneous data from multiple sources, including cleanroom physical parameters, equipment static parameters, real-time operating parameters, and environmental monitoring parameters. The feature extraction module is implemented by a shared encoder composed of KAN layers, responsible for high-dimensional nonlinear feature fusion of the input data to construct a unified latent space representation that can characterize the dynamic characteristics of the system. The multi-task output module consists of multiple parallel KAN sub-networks, corresponding to different control target outputs such as total system air volume, fan frequency, air volume supplied to each room, and valve opening. During the training phase, the model uses gradient backpropagation of the multi-task loss function to collaboratively optimize the shared encoder, enabling it to learn a general feature expression that can simultaneously support cleanliness control, differential pressure stabilization, and energy consumption optimization. In addition, the model has online self-learning capabilities. When the deviation between the predicted value and the actual monitored value remains large or when the preset retraining period is reached, the system will use a fixed-length first-in-first-out data buffer to store effective new data for incremental learning, thereby realizing online adaptive correction of model parameters to track the slow changes in system characteristics.

[0018] Furthermore, the method also includes an operating mode switching step, specifically:

[0019] S0: Determine whether the system needs to enter work mode or duty mode;

[0020] When entering working mode, close the return air and exhaust air switching valve and start the exhaust fan;

[0021] When entering duty mode, the return air and exhaust air switching valve is opened and the exhaust fan is turned off, so that the exhaust air enters the return air system for circulation.

[0022] The operation mode switching step S0 specifically includes:

[0023] S0.1: Predict the mode switching time point based on historical data, production scheduling information, and real-time personnel signals;

[0024] S0.2: When the switching time point is reached, a gradual adjustment strategy is adopted to adjust the total air supply volume, fresh air volume, fan frequency and damper opening in sequence, and the switching valve operation is performed after the system status is stable in order to maintain the pressure difference and cleanliness of each room.

[0025] More specifically, the method also includes an intelligent switching step for system operating modes. The system defines at least two modes: a working mode and a duty mode. In working mode, the return air / exhaust air switching valve is closed and the exhaust fan is started to maintain normal fresh air exchange. In duty mode (such as when no one is producing), the return air / exhaust air switching valve is opened and the exhaust fan is turned off, so that some or all of the air that would otherwise be exhausted to the outside is recirculated back into the air supply system, thereby significantly reducing the demand for fresh air and the related energy consumption for fresh air handling. The timing of mode switching can be predicted based on production plans, historical patterns, and real-time personnel sensor signals. The switching process adopts a gradual adjustment strategy, smoothly transitioning by orderly and coordinated adjustment of the total air volume, fan frequency, and the opening degree of each air valve, avoiding any impact on indoor pressure difference and cleanliness.

[0026] Furthermore, the real-time environmental parameters collected in step S1 include at least: temperature, humidity, pressure difference, particle concentration, and number of personnel. Equipment operating parameters include at least: supply / exhaust fan frequency, supply air valve opening, return / exhaust air valve opening, and real-time energy consumption data of each major piece of equipment.

[0027] Furthermore, in step S3, the air supply volume requirement for each room is determined through the following sub-steps:

[0028] S3.1: Calculate the amount of dust generated in the room based on the number of people and the room area;

[0029] S3.2: Calculate the required number of air changes based on the dust generation and room cleanliness level;

[0030] S3.3: Calculate the required air volume for the room based on the air exchange rate, room area, and net height;

[0031] The cleanroom neural network model corrects the coefficients in the calculation of dust generation and air change rate online based on real-time monitoring data.

[0032] More specifically, to more accurately calculate the room's air supply requirements, step S3 can include a sub-process for calculating air supply volume requirements: First, estimate the real-time dust generation based on the number of people in the room; then, combine the room's cleanliness level requirements (maximum allowable particle concentration) and calculate the required number of air changes using an association model (whose parameters can be corrected online by a neural network); finally, calculate the theoretical air supply volume requirement for the room based on the number of air changes, room area, and net height. This requirement serves as one of the important inputs for optimization calculations.

[0033] Based on the same inventive concept, the present invention also provides a cleanroom air system for implementing the above-described method. The system includes:

[0034] Multiple sensors are deployed in the cleanroom to collect environmental parameters such as temperature, humidity, pressure difference, particle concentration, and number of personnel.

[0035] The ventilation system includes supply fans, exhaust fans, supply air variable volume valves arranged in the supply air branch pipes of each room, return air variable volume valves arranged in the return / exhaust air branch pipes of each room, and return air / exhaust air switching valves connected between the return air main pipe and the exhaust air main pipe.

[0036] Smart meters are used to collect real-time energy consumption data from key operating equipment such as fans and air valves;

[0037] The controller internally stores or has access to the cleanroom's neural network model and intelligent algorithm program. The controller is communicatively connected to all sensors, smart meters, and ventilation system actuators, and is configured to execute any of the aforementioned control methods. Based on the collected data and the model's output, it generates and sends control commands to dynamically adjust the fan frequency and the opening degree of each valve.

[0038] Furthermore, the energy consumption data collected by the smart meter is fed back to the controller in real time, serving as the direct input for calculating the total power consumption of the system in the fitness function of the optimization algorithm, thereby achieving closed-loop optimization with real energy consumption as the target.

[0039] Beneficial effects of the present invention

[0040] Compared with the prior art, the present invention has the following significant advantages:

[0041] Fast dynamic response and high control precision: This invention utilizes a data-driven neural network model to predict system behavior in real time, combined with intelligent algorithms for rolling optimization, enabling rapid response to dynamic disturbances from personnel, equipment, etc. Real-world testing shows that, in the event of sudden changes in operating conditions, the system's recovery time to steady state can be shortened by more than 40% compared to traditional PID control methods.

[0042] Significant global energy-saving optimization: By clearly defining the total system energy consumption as the optimization target and using an optimization algorithm to find the optimal solution within the space that meets the hard constraints of cleanliness and pressure difference in all rooms, a shift from "maintaining parameters" to "optimizing parameters and reducing energy consumption" has been achieved. Combined with exhaust air recirculation technology in the duty mode, the overall system energy consumption can be reduced by more than 30% compared to the traditional fixed air volume operation strategy.

[0043] Highly adaptive and robust: The cleanroom neural network model has online self-learning capabilities, which can continuously fine-tune according to the actual system operation data, automatically adapt to slow time-varying characteristics such as equipment performance degradation and filter resistance changes, as well as differences between different cleanrooms, thereby improving the generalization ability and long-term operational stability of the method.

[0044] Excellent collaborative control effect: The neural network with multi-task learning architecture can naturally learn the coupling relationship between rooms and parameters. Combined with the global search capability of the optimization algorithm, the output control strategy can coordinate the actions of all air valves and fans, effectively avoiding the chain fluctuations caused by local adjustment, and realizing the collaborative stability of environmental parameters in multiple rooms. Attached Figure Description

[0045] Figure 1 This is an overall flowchart of the control method in an embodiment of the present invention. Detailed Implementation

[0046] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. The following embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Example 1: A cleanroom control method using intelligent algorithms

[0047] A cleanroom control method employing intelligent algorithms, characterized by comprising the following steps:

[0048] S1: Collect real-time environmental parameters and equipment operating parameters within the cleanroom;

[0049] S2: Input the environmental parameters and equipment operating parameters into a pre-trained cleanroom neural network model;

[0050] S3: Using cleanliness and pressure difference as constraints and system energy consumption as the optimization objective, the optimal control parameters with the lowest system energy consumption are obtained through intelligent algorithm optimization calculation.

[0051] S4: Dynamically adjust the operating status of the cleanroom ventilation system according to the optimal control parameters.

[0052] Furthermore, the intelligent algorithm used in step S3 is an optimization algorithm. The optimization algorithm inputs the combination of control parameters into the cleanroom neural network model for forward simulation, predicts the system performance indicators, and performs iterative optimization based on a fitness function that includes system energy consumption, pressure difference constraints, and cleanliness constraints.

[0053] More specifically, the workflow includes: taking a combination of various control parameters, including fan frequency and the opening degree of each air valve, as candidate solutions ("particles"), inputting them into the cleanroom neural network model for forward simulation, and quickly predicting the system's total power consumption, pressure difference in each room, and particle concentration under the corresponding scheme; then, evaluating the merits of the candidate solutions based on a fitness function that integrates energy consumption, pressure difference deviation, and cleanliness exceedance; and finally outputting the control parameter combination with the highest fitness, i.e., the system energy consumption is approximately minimized under the premise of strictly satisfying all environmental constraints.

[0054] Furthermore, the cleanroom neural network model is constructed using a data-driven approach and possesses online self-learning capabilities; when the deviation between the predicted value and the measured value exceeds a preset threshold or reaches a preset period, the system adaptively corrects the model parameters based on incremental data.

[0055] Furthermore, the cleanroom neural network model is a deep learning model based on the fusion architecture of KAN network and multi-task learning. Its inputs include cleanroom physical parameters, equipment parameters, real-time operating parameters and environmental parameters, and its outputs include total system air volume, fan frequency and the opening degree of air valves in each room.

[0056] More specifically, the cleanroom neural network model is a data-driven end-to-end deep learning model. Preferably, the model adopts a KAN network and multi-task learning fusion architecture, including an input module, a feature extraction module, and a multi-task output module. The input module receives heterogeneous data from multiple sources, including cleanroom physical parameters, equipment static parameters, real-time operating parameters, and environmental monitoring parameters. The feature extraction module is implemented by a shared encoder composed of KAN layers, responsible for high-dimensional nonlinear feature fusion of the input data to construct a unified latent space representation that can characterize the dynamic characteristics of the system. The multi-task output module consists of multiple parallel KAN sub-networks, corresponding to different control target outputs such as total system air volume, fan frequency, air volume supplied to each room, and valve opening. During the training phase, the model uses gradient backpropagation of the multi-task loss function to collaboratively optimize the shared encoder, enabling it to learn a general feature expression that can simultaneously support cleanliness control, differential pressure stabilization, and energy consumption optimization. In addition, the model has online self-learning capabilities. When the deviation between the predicted value and the actual monitored value remains large or when the preset retraining period is reached, the system will use a fixed-length first-in-first-out data buffer to store effective new data for incremental learning, thereby realizing online adaptive correction of model parameters to track the slow changes in system characteristics.

[0057] Furthermore, the method also includes an operating mode switching step, specifically:

[0058] S0: Determine whether the system needs to enter work mode or duty mode;

[0059] When entering working mode, close the return air and exhaust air switching valve and start the exhaust fan;

[0060] When entering duty mode, the return air and exhaust air switching valve is opened and the exhaust fan is turned off, so that the exhaust air enters the return air system for circulation.

[0061] The operation mode switching step S0 specifically includes:

[0062] S0.1: Predict the mode switching time point based on historical data, production scheduling information, and real-time personnel signals;

[0063] S0.2: When the switching time point is reached, a gradual adjustment strategy is adopted to adjust the total air supply volume, fresh air volume, fan frequency and damper opening in sequence, and the switching valve operation is performed after the system status is stable in order to maintain the pressure difference and cleanliness of each room.

[0064] More specifically, the method also includes an intelligent switching step for system operating modes. The system defines at least two modes: a working mode and a duty mode. In working mode, the return air / exhaust air switching valve is closed and the exhaust fan is started to maintain normal fresh air exchange. In duty mode (such as when no one is producing), the return air / exhaust air switching valve is opened and the exhaust fan is turned off, so that some or all of the air that would otherwise be exhausted to the outside is recirculated back into the air supply system, thereby significantly reducing the demand for fresh air and the related energy consumption for fresh air handling. The timing of mode switching can be predicted based on production plans, historical patterns, and real-time personnel sensor signals. The switching process adopts a gradual adjustment strategy, smoothly transitioning by orderly and coordinated adjustment of the total air volume, fan frequency, and the opening degree of each air valve, avoiding any impact on indoor pressure difference and cleanliness.

[0065] Furthermore, the real-time environmental parameters collected in step S1 include at least: temperature, humidity, pressure difference, particle concentration, and number of personnel. Equipment operating parameters include at least: supply / exhaust fan frequency, supply air valve opening, return / exhaust air valve opening, and real-time energy consumption data of each major piece of equipment.

[0066] Furthermore, in step S3, the air supply volume requirement for each room is determined through the following sub-steps:

[0067] S3.1: Calculate the amount of dust generated in the room based on the number of people and the room area;

[0068] S3.2: Calculate the required number of air changes based on the dust generation and room cleanliness level;

[0069] S3.3: Calculate the required air volume for the room based on the air exchange rate, room area, and net height;

[0070] The cleanroom neural network model corrects the coefficients in the calculation of dust generation and air change rate online based on real-time monitoring data.

[0071] More specifically, to more accurately calculate the room's air supply requirements, step S3 can include a sub-process for calculating air supply volume requirements: First, estimate the real-time dust generation based on the number of people in the room; then, combine the room's cleanliness level requirements (maximum allowable particle concentration) and calculate the required number of air changes using an association model (whose parameters can be corrected online by a neural network); finally, calculate the theoretical air supply volume requirement for the room based on the number of air changes, room area, and net height. This requirement serves as one of the important inputs for optimization calculations.

[0072] Specifically, the control method in this embodiment is implemented according to the following steps:

[0073] System initialization and model loading: When the system starts, the controller loads the pre-trained cleanroom neural network model from the memory and initializes the parameters of the intelligent algorithm (in this example, the particle swarm optimization algorithm PSO) such as the number of particles, the number of iterations, and the inertia weight. At the same time, it reads the fixed physical parameters of the cleanroom (area, net height, and cleanliness level of each room), the rated parameters of the equipment, and the control thresholds (differential pressure setpoint and particle concentration limit).

[0074] Data Acquisition (S1): The system enters a loop operation. At the beginning of each control cycle (e.g., set to 5 minutes), the controller acquires real-time data from various sensors via an industrial bus or IoT gateway:

[0075] Environmental parameters: Temperature and humidity sensors, differential pressure sensors, online particle monitors, and infrared personnel counter data for each cleanroom.

[0076] Equipment operating parameters: frequency feedback of the inverters for the supply and exhaust fans, and valve position opening feedback values ​​of each supply air variable volume valve and return / exhaust air variable volume valve.

[0077] Energy consumption data: The total power consumption or cumulative power increment of the blower, exhaust fan and valve actuators uploaded by the smart meter.

[0078] Operating Mode Judgment and Switching (Optional Step S0): The controller, considering the current time, production plan information, and real-time personnel numbers, determines the appropriate operating mode. If a switch from work mode to shift mode is required, a gradual switching procedure is initiated: First, the total system air volume setpoint is gradually reduced, correspondingly decreasing the air supply fan frequency; simultaneously, the fresh air valve opening is proportionally reduced, gradually lowering the fresh air volume to the level required to maintain only the necessary differential pressure; next, the exhaust fan is shut off; after the system air volume stabilizes at a lower level, the return and exhaust air switching valves are opened to complete the mode switch. During this process, the controller counteracts differential pressure disturbances caused by changes in total air volume by fine-tuning the air valves in each room.

[0079] Air volume demand calculation (S3 sub-step): For each room, the controller, based on the currently collected data on the number of people (P) and room area (A), in square meters, substitutes these values ​​into the formula.

[0080] Estimate the real-time dust emission rate G, in units of pc / (m³·min).

[0081] Wherein, coefficient C (dust generation per person, per person / (min·person)) is initially set to... The initial value of the particle concentration monitoring data can be set based on experience and corrected online by the neural network model during system operation based on the correlation between the actual number of personnel and the particle concentration monitoring data.

[0082] Then, based on the room's cleanliness level (corresponding to the maximum permissible particle concentration N, unit: pc / m³) and dust generation rate G, unit: pc / (m³·min), the formula is used to...

[0083] Calculate the required air exchange rate n, times / hour. The coefficients a and b in the formula are initially set as empirical constants (a, coefficient, initial value 4566.3; b, coefficient, initial value 55.264), which are also learned and corrected online by the neural network model based on the statistical relationship between the air exchange rate and the final cleanliness achieved in historical data.

[0084] Finally, based on the air exchange rate n (times / hour), room area A (㎡), and net height h (m), the theoretical air supply volume requirement for the room is calculated. This requirement value serves as an important reference boundary for subsequent optimization. The formula is as follows:

[0085] Q, Cleanroom air supply volume, m³ / h

[0086] Neural Network Prediction and Intelligent Optimization (S2, S3):

[0087] Model input: The real-time data collected in step S2, the demand data calculated in step S4, and the system physical parameters are used together to construct the neural network input vector for the cost control cycle.

[0088] Algorithm Optimization: The Particle Swarm Optimization (PSO) algorithm begins operation. The algorithm initializes a swarm of "particles," each representing a possible control scheme, such as a specific set of supply / exhaust fan frequencies and valve opening values. For each particle, the controller inputs its represented control parameters into the loaded cleanroom neural network model for forward inference (simulation). The model quickly outputs prediction results: the total system power consumption P, and the predicted pressure difference for each room. and predicting particle concentration .

[0089] Fitness evaluation: Based on the prediction results, calculate the fitness value F of the particle. The fitness function is designed as follows:

[0090] in,

[0091] Used to penalize differential pressure deviations exceeding the set value and thus exceeding the permissible deviation. The situation.

[0092] Used to punish situations where particle concentration exceeds the upper limit.

[0093] The constraint function for the pressure difference in the i-th room;

[0094] The constraint function for the cleanliness of the i-th room;

[0095] The pressure difference in the i-th room predicted by the neural network model;

[0096] The set value of the pressure difference in the i-th room;

[0097] The concentration of particles of a specific size in the i-th room predicted by the neural network model;

[0098] The allowable particle concentration limit for the i-th room.

[0099] , , This is a weighting coefficient used to balance the importance of energy consumption with the constraints of different room environments. The larger the fitness value F, the lower the energy consumption of the scheme while meeting the constraints.

[0100] Iterative Update: The PSO algorithm updates the velocity and position of all particles based on their fitness values ​​(i.e., explores new combinations of control parameters) and begins the next generation of evaluation. After a preset number of iterations, the algorithm converges, selecting the particle with the highest fitness; the control parameter combination represented by this particle is the optimal solution found in this control cycle.

[0101] Control command issuance and execution (S4): The controller converts the optimal control parameters obtained through optimization, namely the target frequency of the supply / exhaust fans and the target opening degree of each supply / return / exhaust variable air volume valve, into specific control commands, which are then issued to the corresponding frequency converters and electric valves through the actuators. The equipment operates according to the new commands, changing the supply air volume, return air volume, and exhaust air volume, thereby achieving dynamic adjustment of the cleanroom environmental parameters.

[0102] Online self-learning model: During operation, the system continuously compares the actual monitored environmental parameters (pressure difference, particle concentration) and equipment energy consumption data with the predictions previously made by the neural network model. When the prediction error exceeds a preset threshold for several consecutive periods, or reaches a fixed retraining cycle (e.g., one week), the system automatically initiates the online learning program. At this time, the system retrieves valid normal operating condition data from the most recent period (e.g., one month) from the first-in-first-out data buffer and incrementally trains the neural network model, updating its network weights. This allows the model to continuously adapt to changes in system characteristics and maintain prediction accuracy. Example 2: A cleanroom air system employing intelligent algorithms

[0103] The present invention also provides a cleanroom air system for implementing the above method. The system includes:

[0104] Multiple sensors are deployed in the cleanroom to collect environmental parameters such as temperature, humidity, pressure difference, particle concentration, and number of personnel.

[0105] The ventilation system includes supply fans, exhaust fans, supply air variable volume valves arranged in the supply air branch pipes of each room, return air variable volume valves arranged in the return / exhaust air branch pipes of each room, and return air / exhaust air switching valves connected between the return air main pipe and the exhaust air main pipe.

[0106] Smart meters are used to collect real-time energy consumption data from key operating equipment such as fans and air valves;

[0107] The controller internally stores or has access to the cleanroom's neural network model and intelligent algorithm program. The controller is communicatively connected to all sensors, smart meters, and ventilation system actuators, and is configured to execute any of the aforementioned control methods. Based on the collected data and the model's output, it generates and sends control commands to dynamically adjust the fan frequency and the opening degree of each valve.

[0108] Furthermore, the energy consumption data collected by the smart meter is fed back to the controller in real time, serving as the direct input for calculating the total power consumption of the system in the fitness function of the optimization algorithm, thereby achieving closed-loop optimization with real energy consumption as the target.

[0109] More specifically:

[0110] Reference Figure 1 The system in this embodiment is used to implement the method described in Embodiment 1. The system includes:

[0111] Sensing layer: Temperature and humidity sensors, differential pressure sensors, online particle monitors, and infrared sensors are installed in each key room of the cleanroom. Air volume meters can also be installed at key locations in the air ducts.

[0112] The control layer includes the supply fan (variable frequency control) inside the air conditioning unit, the independent exhaust fan (variable frequency control), the supply air variable volume valve on each room's supply air branch duct, the return / exhaust air variable volume valve on each room's return air / exhaust air branch duct, and the electrically operated return / exhaust switching valve installed between the return air main duct and the exhaust air main duct. All fans and valves are equipped with actuators and feedback devices.

[0113] Metering layer: Smart meters are installed in the power distribution cabinet to collect the total power consumption data of the supply fan, exhaust fan and valves of the entire air system in real time.

[0114] Control Layer: The core is an industrial-grade controller (such as an industrial PC or high-performance PLC). This controller has a built-in or connected database that stores historical operating data, production schedules, and neural network models. The controller connects to all devices in the sensing, metering, and execution layers via communication networks (such as Ethernet, BACnet, Modbus). The controller runs the complete control algorithm program described in Example 1, which is responsible for data aggregation, model invocation, optimization calculation, and generating control commands.

[0115] When the system is working, the controller acts like an "intelligent brain," constantly sensing changes in the environment, "thinking" about the consequences of different control strategies through neural networks, and using optimization algorithms to "decide" the best course of action, directing the fans and valves to execute, thereby achieving efficient, energy-saving, and intelligent management and control of the cleanroom environment.

[0116] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A clean room control method using an intelligent algorithm, characterized by, Includes the following steps: S1: Collect real-time environmental parameters and equipment operating parameters within the cleanroom; S2: Input the environmental parameters and equipment operating parameters into a pre-trained cleanroom neural network model; S3: Using cleanliness and pressure difference as constraints and system energy consumption as the optimization objective, the optimal control parameters with the lowest system energy consumption are obtained through intelligent algorithm optimization calculation. S4: Dynamically adjust the operating status of the cleanroom ventilation system according to the optimal control parameters.

2. The method of claim 1, wherein, The intelligent algorithm used in step S3 is an optimization algorithm. The optimization algorithm inputs the combination of control parameters into the cleanroom neural network model for forward simulation, predicts the system performance indicators, and performs iterative optimization based on a fitness function that includes system energy consumption, pressure difference constraints, and cleanliness constraints.

3. The method according to claim 1 or 2, characterized in that, The cleanroom neural network model is constructed using a data-driven approach and has online self-learning capabilities. When the deviation between the predicted value and the measured value exceeds a preset threshold or reaches a preset period, the system adaptively corrects the model parameters based on incremental data.

4. The method of claim 1, wherein, It also includes the steps for switching operating modes, specifically: S0: Determine whether the system needs to enter work mode or duty mode; When entering working mode, close the return air and exhaust air switching valve and start the exhaust fan; When entering duty mode, the return air and exhaust air switching valve is opened and the exhaust fan is turned off, so that the exhaust air enters the return air system for circulation.

5. The method according to claim 4, characterized in that, The operation mode switching step S0 specifically includes: S0.1: Predict the mode switching time point based on historical data, production scheduling information, and real-time personnel signals; S0.2: When the switching time point is reached, a gradual adjustment strategy is adopted to adjust the total air supply volume, fresh air volume, fan frequency and damper opening in sequence, and the switching valve operation is performed after the system status is stable in order to maintain the pressure difference and cleanliness of each room.

6. The method according to claim 1, characterized in that, The real-time environmental parameters collected in step S1 include: temperature, humidity, pressure difference, particle concentration, and number of personnel; the equipment operating parameters include: supply fan frequency, exhaust fan frequency, supply air valve opening, return / exhaust air valve opening, and real-time energy consumption of the equipment.

7. The method according to claim 2, characterized in that, The cleanroom neural network model is a deep learning model based on the fusion architecture of KAN network and multi-task learning. Its inputs include cleanroom physical parameters, equipment parameters, real-time operating parameters and environmental parameters. Its outputs include total system air volume, fan frequency and the opening degree of air valves in each room.

8. The method according to claim 1, characterized in that, In step S3, the air supply volume requirement for each room is determined through the following sub-steps: S3.1: Calculate the amount of dust generated in the room based on the number of people and the room area; S3.2: Calculate the required number of air changes based on the dust generation and room cleanliness level; S3.3: Calculate the required air volume for the room based on the air exchange rate, room area, and net height; The cleanroom neural network model corrects the coefficients in the calculation of dust generation and air change rate online based on real-time monitoring data.

9. A cleanroom air system employing intelligent algorithms, characterized in that, include: Multiple sensors are used to collect environmental parameters and the number of people in the cleanroom; Supply fan, exhaust fan, supply air variable air volume valve, return / exhaust air variable air volume valve and return / exhaust switching valve; A controller configured to perform the control method as described in any one of claims 1-8, wherein the controller generates control commands to dynamically adjust the fan and the air valve based on the data collected by the sensor and the output of the cleanroom neural network model.

10. The cleanroom ventilation system according to claim 9, characterized in that, It also includes smart meters, which are used to collect real-time energy consumption data of the fan and the valve and feed it back to the controller so that the controller can evaluate and optimize the system energy efficiency in optimization calculations.

Images