Control method and device for fresh air system of clean room, medium and computer program

Abstract

The invention provides a control method and device for a fresh air system of a clean room, a medium and a computer program, and relates to the technical field of clean room ventilation control. The method comprises the steps that a system collects air quality data of different detection points of a clean room in real time, and an original data sequence is generated; key feature data is extracted from the original data sequence, and dimensionality reduction is performed on the key feature data by using PCA to form a dimensionality reduction feature matrix; respectively inputting the dimension reduction feature matrix into ARIMA and LSTM to obtain prediction results, and performing weighted fusion to obtain a fusion prediction result of the air quality; and based on the fusion prediction result and the detection point position, marking a potential pollution source position, obtaining a diffusion path simulation result according to the diffusion model, and based on the fusion prediction result, the potential pollution source position and the diffusion path simulation result, generating a real-time dynamic control strategy to control the operation state of the fresh air system. Efficient and energy-saving operation can be achieved while it is ensured that the air quality reaches the standard.

Description

Technical Field

[0001] The present invention relates to the technical field of clean room ventilation control, and in particular to a control method, equipment, medium and computer program for a clean room fresh air system. Background Art

[0002] In semiconductor manufacturing, cleanroom air quality is crucial. Tiny particles or chemical contaminants can cause irreversible damage to chip processing, directly impacting product yield and performance. Therefore, semiconductor cleanrooms must maintain extremely high levels of cleanliness. However, existing air quality management methods primarily rely on a fixed fresh air volume, monitoring air quality based on preset thresholds and triggering control measures when these thresholds are exceeded. This approach lacks foresight regarding air quality changes and often fails to respond promptly to sudden pollution events, resulting in air quality failing to meet standards for a short period of time, thus impacting product manufacturing stability.

[0003] Furthermore, existing systems lack the means to accurately locate and control pollution sources. When contamination occurs within a cleanroom, the system typically addresses it by increasing fresh air volume and filtering frequency, but this approach is inefficient. Due to the complex cleanroom environment and the potentially diverse distribution of pollution sources, traditional methods struggle to achieve pinpoint control of pollution sources, often causing cleanroom systems to operate at high loads, increasing energy consumption and operating costs. These issues make it difficult for existing air quality management systems to achieve efficient and energy-efficient operation while ensuring air quality meets standards. Summary of the Invention

[0004] Embodiments of the present invention provide a method, device, medium, and computer program for controlling a clean room fresh air system to solve the problem of achieving efficient and energy-saving operation while ensuring that air quality meets standards.

[0005] In a first aspect, an embodiment of the present invention provides a method for controlling a clean room fresh air system, characterized by comprising: The system collects air quality data from different detection points in the clean room in real time and generates a raw data sequence. The air quality data includes: particulate matter concentration, total acidic pollutant TA concentration, total alkaline pollutant TB concentration, volatile organic compound (VOC) concentration, temperature, humidity, wind speed and pressure difference. Extracting key feature data from the original data sequence, and performing dimensionality reduction on the key feature data using principal component analysis (PCA) to form a reduced-dimensional feature matrix; The reduced-dimensional feature matrix is ​​input into an autoregressive integrated moving average model (ARIMA) and a long short-term memory network (LSTM) to obtain prediction results, and weighted fusion is performed to obtain a fused prediction result of air quality; Based on the fusion prediction results and the detection point locations, the potential pollution source locations are marked, and the diffusion path simulation results are obtained according to the diffusion model. Based on the fusion prediction results, the potential pollution source locations and the diffusion path simulation results, a real-time dynamic control strategy is generated to adjust the operating status of the fresh air system.

[0006] In a possible implementation, marking the location of a potential pollution source based on the fusion prediction result and the detection point location includes: Determine the pollution source area through spatial gradient analysis based on the fusion prediction results and the detection point locations; The Gaussian mixture model is used to establish the source point characteristic distribution. The pollutant concentration and detection point location of the detection point are used as input to calculate the probability that each detection point location belongs to the source point cluster. When the probability exceeds the preset threshold, the detection point location is marked as a potential pollution source location.

[0007] In a possible implementation, determining the pollution source area through spatial gradient analysis based on the fusion prediction result and the detection point location includes: The pollutant concentrations at each detection point in the fusion prediction results are interpolated into a continuous distribution, and the gradient and Laplace value are calculated based on the interpolation results to determine the spatial variation trend; The grids with local peaks marked and the changing trend radiating outward are candidate sources; For each candidate source, the wind speed and pressure difference are combined to trace back multiple times along the wind direction and negative gradient direction, and the pollution source area is determined using the change trend threshold, peak threshold and consistency threshold.

[0008] In one possible implementation, obtaining a diffusion path simulation result according to a diffusion model includes: Calculate the predicted pollutant concentration at each detection point at different time points based on the diffusion model and construct a predicted pollutant concentration sequence; The diffusion path simulation results are obtained based on the predicted pollutant concentration sequence at each detection point; Wherein, the diffusion model formula is:

[0009] , Indicates location In time The predicted pollutant concentration, is the initial pollutant concentration, is the diffusion coefficient, To the location of the pollution source distance.

[0010] In a possible implementation, before marking the location of the potential pollution source based on the fused prediction result and the detection point location, the method further includes: Determine, based on the fusion prediction results, whether there are indicators among the particulate matter concentration, the total acidic pollutant TA concentration, the total alkaline pollutant TB concentration, and the VOC concentration whose concentration increases beyond a set threshold; Correspondingly, when there is an indicator that the concentration increases beyond the set threshold, the potential pollution source location is marked based on the fusion prediction result and the detection point location and subsequent operations are performed; otherwise, the fresh air system maintains the current state unchanged.

[0011] In a possible implementation, the real-time dynamic control strategy includes: a target fresh air volume, a fan filter unit (FFU) adjustment range, and an FFU wind speed adjustment amount.

[0012] In a possible implementation, after generating the real-time dynamic control strategy, the method further includes: Update and obtain air quality data, energy consumption of fresh air systems and purification equipment; The real-time collected air quality data, the energy consumption of the fresh air system and purification equipment are input into the trained deep reinforcement learning network to obtain the optimal control strategy of the fresh air system; The deep reinforcement learning network training process is as follows: Design the state space S of deep reinforcement learning, which includes the real-time air quality parameters of the clean room, the control parameters of the real-time dynamic control strategy, and the system energy consumption data; Set the action space A, which includes the action mode a of increasing or decreasing the fresh air volume, adjusting the FFU wind speed, and switching the FFU switch state; Constructing the reward function:

[0013] in, Energy consumption of fresh air system and purification equipment; 、 is the weight coefficient of the reward function, which represents the weight of the energy consumption and air quality parameters of the fresh air system and purification equipment in the reward. is the air quality compliance rate; A deep reinforcement learning framework based on the proximal policy optimization (PPO) algorithm was built. S, A, and R were extracted based on historical cleanroom data, and the deep reinforcement learning framework was trained to obtain a trained deep reinforcement learning network. The cleanroom historical data included air quality parameters, energy consumption of the fresh air system and purification equipment, target fresh air volume, FFU adjustment range, and FFU wind speed adjustment.

[0014] In a second aspect, an embodiment of the present invention provides a control device for a clean room fresh air system, comprising: The acquisition module is used to collect air quality data from different detection points in the clean room in real time and generate a raw data sequence; wherein the air quality data includes: particulate matter concentration, total acid pollutant TA concentration, total alkaline pollutant TB concentration, VOC concentration, temperature, humidity, wind speed and pressure difference; A dimensionality reduction module is used to extract key feature data from the original data sequence, and use PCA to reduce the dimension of the key feature data to form a reduced dimension feature matrix; A fusion prediction module is used to input the reduced dimension feature matrix into ARIMA and LSTM respectively to obtain prediction results, and perform weighted fusion to obtain a fusion prediction result of air quality; The fresh air system control module is used to mark the location of potential pollution sources based on the fusion prediction results and detection point locations, and obtain diffusion path simulation results according to the diffusion model. Based on the fusion prediction results, potential pollution source locations and diffusion path simulation results, a real-time dynamic control strategy is generated to control the operating status of the fresh air system.

[0015] In a third aspect, an embodiment of the present invention provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method in the first aspect or any possible implementation of the first aspect is implemented.

[0016] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method in the first aspect or any possible implementation of the first aspect.

[0017] In a fifth aspect, an embodiment of the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the method in the first aspect or any possible implementation of the first aspect.

[0018] In an embodiment of the present invention, multi-dimensional air quality data of the clean room is first collected in real time to ensure comprehensive monitoring of the air quality status and avoid the risk of pollution missed by a single indicator. Then, the dimensionality of the key feature data is reduced by PCA, which effectively reduces the amount of subsequent model calculations and reduces the system operation load; the dimensionality-reduced data is then input into ARIMA and LSTM and weighted fused, combining ARIMA's ability to capture periodic changes with LSTM's adaptability to short-term fluctuations to improve the accuracy of air quality prediction and achieve forward-looking judgment on air quality changes. Finally, a control strategy is generated based on the fusion prediction results, pollution source location, and diffusion path simulation to make the fresh air system control more targeted and avoid the lag of the traditional fixed fresh air volume. This can not only respond to pollution risks in a timely manner and ensure the stability of clean room air quality, but also reduce unnecessary energy waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is an application scenario diagram of the clean room fresh air system control method provided by an embodiment of the present invention; Figure 2 This is a flow chart of the implementation of the control method for the clean room fresh air system provided by an embodiment of the present invention; Figure 3 Schematic diagram of the structure of a control device for a clean room fresh air system provided by an embodiment of the present invention; Figure 4 is a schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] This invention is applicable to a variety of fields requiring a highly clean air environment, particularly industries that are extremely sensitive to minute contaminants, such as semiconductor manufacturing, biomedicine, precision instrument production, and aerospace. In these fields, the stability and cleanliness of cleanroom air quality directly impact product quality and production efficiency. This invention provides an intelligent and efficient air quality management solution, effectively improving production yields and reducing operating costs.

[0021] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] Figure 1 This is an application scenario diagram of the method for controlling the fresh air system in a clean room provided by an embodiment of the present invention. Figure 1 As shown in the figure, multiple concentration sensors, FFUs, fresh air equipment, and ambient air exhaust vents are distributed in the clean room. Among them, FFUs, fresh air equipment, and ambient air exhaust vents constitute a complete fresh air system.

[0023] The main purpose of regulating the fresh air volume is to reduce the concentration of pollutants in the clean room by diluting them. Based on the predicted concentration of pollutants, the system can dynamically adjust the fresh air volume.

[0024] In actual implementation, this involves adjusting the operating state of the air ducts, primarily by controlling the operating states of the FFUs and ambient exhaust vents to optimize airflow direction. If the contamination source is located near a particular FFU, the air velocity in that area can be reduced while increasing the air velocity of surrounding FFUs. The ambient exhaust vents can also be opened to direct contaminants toward the vents, preventing them from spreading to other areas of the cleanroom.

[0025] The control terminal is used to remotely collect concentration detection data from the concentration sensor, operating parameters and energy consumption data of the FFU and fresh air equipment, so as to achieve efficient and energy-saving operation while ensuring that the air quality meets the standards.

[0026] The sensors in this application include particulate matter concentration, total acidic pollutant TA concentration, total alkaline pollutant TB concentration, VOC concentration, temperature, humidity, wind speed and pressure difference sensors.

[0027] To comprehensively monitor cleanroom air quality, this solution requires the collection of multiple air quality indicators. Particulate matter concentrations (PM) primarily include PM0.5, PM0.3, PM0.2, and PM0.1, reflecting the concentration of particles with diameters less than 0.5, 0.3, 0.2, and 0.1 microns. TA concentration represents the total concentration of acidic pollutants in the cleanroom air, primarily including hydrogen fluoride (HF), hydrogen chloride (HCl), and sulfur dioxide (SO2). TB concentration represents the total concentration of alkaline pollutants in the cleanroom air, primarily including ammonia (NH3), organic amines (such as triethylamine and dimethylethanolamine), and alkylamine compounds. VOC concentration primarily identifies potential chemical pollutants in the air.

[0028] In addition, temperature and humidity measurements are used to determine the operating status of air conditioning systems and are also correlated with pollution diffusion. Wind speed and pressure differential measurements reflect air flow and pressure differentials, providing airflow distribution information to the control system.

[0029] Sensors are arranged in different areas of the clean room to ensure data coverage and high-precision collection in key areas.

[0030] Installed at the fresh air inlet and outlet, these sensors monitor air quality entering and exiting the cleanroom, capturing the degree of air contamination both inside and out. Sensors can be placed in key internal areas, such as those with high traffic volume or concentrated equipment, to monitor local air quality fluctuations. Distributed sensors can be installed in the four corners and central area of ​​the cleanroom to monitor air quality uniformity and contamination dispersion.

[0031] In the specific implementation process, these sensors are connected to the Internet of Things (IoT) module and transmit data to edge computing nodes in real time to provide support for data processing and subsequent analysis.

[0032] To ensure real-time data collection, the system defaults to collecting data every 10 seconds. This is the baseline sampling frequency under stable conditions and sufficient to provide real-time data support. Furthermore, the system monitors air quality fluctuations in real time and automatically adjusts the sampling frequency based on the magnitude of fluctuations. For example, if significant fluctuations in air quality indicators (such as PM, TA, and TB) are detected, the sampling frequency will be automatically increased to every 2 seconds to respond to sudden pollution incidents and ensure that the spread of pollutants within the cleanroom is captured promptly.

[0033] Dynamic sampling frequency formula:

[0034] in, is the current sampling frequency, is the initial sampling frequency (10 seconds), is the adjustment coefficient, is the rate of change of air quality (such as the rate of change of PM). When it is larger, the system automatically increases the sampling frequency.

[0035] Before implementing the cleanroom fresh air system control method provided in this application, sensor detection data needs to be preprocessed. Optionally, the data collected by the sensors is transmitted to an edge computing node via an Internet of Things module for data preprocessing to improve data processing efficiency. Preprocessing includes missing value processing, outlier removal, and noise processing.

[0036] In the process of missing value processing, linear interpolation is used to fill in short-term missing data. The formula is as follows:

[0037] During the outlier elimination process, the triple standard deviation method is used to eliminate outliers to ensure data reliability: Marked as an exception.

[0038] During the noise processing, the sliding average method is used to smooth the data to reduce the short-term random fluctuations that may be generated by the sensor. The sliding average formula is:

[0039] in, is the smoothed data, is the window size, usually 3 to 5, is the original data point in the window.

[0040] The control method of the clean room fresh air system provided in the embodiment of the present application is developed on the above basis. In order to make the specific process of the control method of the clean room fresh air system clearer, the method embodiment is described in detail below with reference to the accompanying drawings.

[0041] See also Figure 2 , which shows a flow chart of the implementation of the control method of the clean room fresh air system provided by an embodiment of the present invention, including the following steps: In step S201, the system collects air quality data from different detection points in the clean room in real time and generates a raw data sequence. The air quality data includes: particulate matter concentration, total acidic pollutant TA concentration, total alkaline pollutant TB concentration, VOC concentration, temperature, humidity, wind speed and pressure difference.

[0042] The execution subject of each embodiment of this application can be a server, processor, microprocessor or other device with data processing function. In the actual implementation process, the specific implementation method of the execution subject can be selected according to actual needs. This embodiment does not impose any special restrictions on this, as long as it is a device with data processing function. Figure 1 The control terminal shown in the figure is used as the execution subject for explanation.

[0043] In the specific implementation process, the specific classification of air quality data can be found in the above Figure 1 Applicable to specific application fields, according to the different compositions of gases and particulate matter in the clean room environment, the corresponding detection classifications of particulate matter concentration, total acid pollutant TA concentration, total alkaline pollutant TB concentration, and VOC concentration are different.

[0044] In order to improve the regulation efficiency of the fresh air system in the clean room and implement corresponding indoor environment changes, the control terminal collects the air quality data of different detection points in the clean room and generates the original data sequence recorded as ,Will As input for feature extraction and dimensionality reduction.

[0045] S202, extracting key feature data from the original data sequence, and using PCA to reduce the dimension of the key feature data to form a reduced-dimensional feature matrix.

[0046] From the original data The key features are extracted from the data to simplify the data dimension and highlight the core parameters that affect air quality changes.

[0047] The key features extracted are those closely related to changes in air quality, such as the change rates of PM, TA, TB, and the fluctuation range of temperature and humidity. . It contains key air quality parameters, providing concise input for subsequent forecasts.

[0048] In order to facilitate analysis and processing, PCA is used to analyze the feature data. Perform dimensionality reduction to form the final dimensionality reduction feature matrix . PCA formula:

[0049] in, is the eigenvector matrix, yes The transposed matrix of It is the feature matrix after dimensionality reduction, representing the simplified core air quality parameters.

[0050] Dimensionality reduction data This includes essential air quality information, reducing computational complexity and data redundancy. The prediction model only needs to process a small number of core features, enabling more efficient prediction of air quality trends and improved model accuracy.

[0051] S203: Input the reduced dimension feature matrix into ARIMA and LSTM respectively to obtain prediction results, and perform weighted fusion to obtain a fused prediction result of air quality.

[0052] The reduced dimension feature matrix is ​​input into ARIMA and LSTM respectively, aiming to utilize the reduced dimension data Predicting future air quality trends provides a basis for proactively adjusting control strategies. The ARIMA model captures cyclical changes in air quality (such as the diurnal cycle), providing a foundation for short-term forecasts. The LSTM model is primarily used to capture short-term fluctuations in air quality, improving forecast accuracy.

[0053] Among them, the ARIMA formula is:

[0054] in, is the predicted value of current air quality; for The predicted value of air quality corresponding to the moment; is the autoregressive coefficient, responsible for capturing the trend of the data; is the coefficient of the moving average (MA) part, which is used to process noise; is the current white noise term, i.e., random error; for The white noise term corresponding to the moment.

[0055] The LSTM calculation steps include: Forget Gate:

[0056] Input Gate:

[0057] Status Update:

[0058] Output Gate:

[0059] in, Output vector for the forget gate; is the sigmoid activation function; is the forget gate weight matrix; is the hidden state vector of LSTM at the previous moment; is the dimension-reduced feature matrix; is the forget gate bias term; Output vector for the input gate; is the input gate bias term; is the cell state vector at the current moment; is the cell state vector at the previous moment; is the candidate cell state vector; is the hyperbolic tangent activation function; The hidden state vector at the current moment; Output vector for the output gate.

[0060] Finally, the outputs of ARIMA and LSTM are weighted and fused to obtain the final prediction result. .

[0061] The fusion formula is:

[0062] in, and is the prediction output of ARIMA and LSTM, and is the weight coefficient.

[0063] S204: Based on the fusion prediction results and the detection point locations, the potential pollution source locations are marked, and the diffusion path simulation results are obtained according to the diffusion model. Based on the fusion prediction results, the potential pollution source locations and the diffusion path simulation results, a real-time dynamic control strategy is generated to control the operating status of the fresh air system.

[0064] In the specific implementation process, based on the prediction results The air quality change characteristics in the air are combined with the sensor location data (corresponding to the detection point location) to mark the location of potential pollution sources. , based on the prediction results , location of pollution source and diffusion path simulation , generating real-time dynamic control strategies to adjust the operating status of the fresh air system. The control system automatically increases the fresh air volume to dilute pollutants, and adjusts the FFU status to adjust the pollutant diffusion path, while taking into account the overall energy consumption of the fresh air system.

[0065] In this embodiment, multi-dimensional air quality data of the clean room is first collected in real time to ensure comprehensive monitoring of the air quality status and avoid the risk of pollution missed by a single indicator. Then, the dimensionality of the key feature data is reduced through PCA, which effectively reduces the amount of subsequent model calculations and reduces the system operation load; the dimensionality-reduced data is then input into ARIMA and LSTM and weighted fused, combining ARIMA's ability to capture periodic changes with LSTM's adaptability to short-term fluctuations to improve the accuracy of air quality prediction and achieve forward-looking judgment on air quality changes. Finally, a control strategy is generated based on the fusion prediction results, pollution source location, and diffusion path simulation to make the fresh air system control more targeted and avoid the lag of the traditional fixed fresh air volume. It can not only respond to pollution risks in a timely manner and ensure the stability of clean room air quality, but also reduce unnecessary energy waste. In one possible implementation, based on the fusion prediction results and the detection point locations, the potential pollution source locations are marked, including: Based on the fusion prediction results and the location of the detection points, the pollution source area is determined through spatial gradient analysis; The Gaussian mixture model is used to establish the source point characteristic distribution. The pollutant concentration and detection point location of the detection point are used as input to calculate the probability that each detection point location belongs to the source point cluster. When the probability exceeds the preset threshold, the detection point location is marked as a potential pollution source location.

[0066] In this example, pollution source areas are initially delineated through spatial gradient analysis, combined with fusion prediction results and detection point locations, providing a spatial basis for locating potential pollution sources. A Gaussian mixture model is then used to construct a source point distribution using the pollutant concentration and location of the detection points as input. By calculating the probability of each location belonging to a source point cluster and combining it with a preset threshold to label potential pollution sources, pollution source location is upgraded from "regional determination" to "probabilistic quantification." This effectively reduces the risk of misjudgment caused by single-space analysis, makes potential pollution source location marking more accurate and reliable, and provides a clearer target for the generation of targeted control strategies.

[0067] The above embodiment summarizes the process of marking potential pollution source locations based on fused prediction results and detection point locations, which consists of two steps: determining the pollution source area and marking the pollution source location. The following describes the determination of the pollution source area through spatial gradient analysis.

[0068] In one possible implementation, based on the fused prediction results and the locations of the detection points, the pollution source area is determined through spatial gradient analysis, including: The pollutant concentrations at each detection point in the fusion prediction results are interpolated into a continuous distribution, and the gradient and Laplace value are calculated based on the interpolation results to determine the spatial variation trend; The grids with local peaks marked and the changing trend radiating outward are candidate sources; For each candidate source, the wind speed and pressure difference are combined to trace back multiple times along the wind direction and negative gradient direction, and the pollution source area is determined using the change trend threshold, peak threshold and consistency threshold.

[0069] In the actual implementation process, the fusion prediction results include at least the pollutant concentration at the detection point, the coordinates of the detection point, the wind speed / direction and pressure difference at the same time; optionally, environmental quantities such as temperature and humidity are also included.

[0070] Before using the Gaussian mixture model to determine the location of the potential pollution source, the pollution source area to which the potential pollution source belongs is first determined.

[0071] In this embodiment, by refining the spatial gradient analysis process, the pollutant concentrations at discrete detection points are first interpolated into a continuous distribution to fill the data blind spots in areas without detection points, ensuring a complete perception of the overall concentration field in the clean room; then, the spatial change trend is clarified by calculating the gradient and Laplace value, and the grids with local peaks and outward radiating change trends are accurately identified as candidate sources, and areas with pollution diffusion characteristics are preliminarily screened out; finally, the wind speed and pressure difference are combined to trace back multiple time data along a specific direction, and through comprehensive judgment of multiple thresholds, interference factors such as temporary pollution residues are effectively eliminated, further improving the precision and accuracy of locating the pollution source area and improving the accuracy of the control strategy.

[0072] In one possible implementation, obtaining a diffusion path simulation result according to a diffusion model includes: Calculate the predicted pollutant concentration at each detection point at different time points based on the diffusion model and construct a predicted pollutant concentration sequence; The diffusion path simulation results are obtained based on the predicted pollutant concentration sequence at each detection point; Among them, the diffusion model formula is:

[0073] , Indicates location In time The predicted pollutant concentrations, is the initial pollutant concentration, is the diffusion coefficient, To the location of the pollution source distance.

[0074] In this embodiment, a diffusion model is used to calculate the predicted pollutant concentrations at each detection point at different times and construct a concentration sequence. The relationship between initial concentration, diffusion coefficient, and distance in the model formula is then used to quantitatively predict the changes in pollutant concentration at different locations at different times, thereby clearly reconstructing the pollution diffusion path and range. Compared to traditional methods that rely solely on threshold monitoring, this method can proactively identify pollution diffusion trends, allowing the fresh air system control to pre-emptively plan diffusion paths, preventing pollution from spreading to critical areas, and further improving the proactive and effective control of cleanroom air quality.

[0075] In one possible implementation, before marking the location of the potential pollution source based on the fused prediction result and the detection point location, the following is further included: Based on the fusion prediction results, determine whether there are indicators among the particulate matter concentration, total acidic pollutant TA concentration, total alkaline pollutant TB concentration, and VOC concentration whose concentration exceeds the set threshold; Accordingly, when there is an indicator that the concentration increases beyond the set threshold, the potential pollution source location is marked based on the fusion prediction results and the detection point location and subsequent operations; otherwise, the fresh air system maintains its current state.

[0076] The main purpose of adjusting the fresh air system is to reduce the pollution concentration in the clean room by diluting the pollutants through fresh air volume adjustment. Based on the predicted results of the pollutant concentration, the system can dynamically adjust the operating status of the fresh air system. The rules for adjusting the operating status of the fresh air system are: If the prediction result If the air quality forecast indicates a deterioration in the future (e.g., increased PM or VOC concentrations), the system automatically increases the fresh air volume to dilute the pollutants. If the air quality forecast is relatively stable, the system maintains the base fresh air volume to reduce energy consumption.

[0077] In this embodiment, the method first determines whether core pollution indicators exceed a threshold before locating the pollution source. Subsequent location and control operations are initiated only when pollution risk exists. If the indicators are stable, the fresh air system maintains its current state. This avoids the problem of traditional systems constantly executing complex control processes regardless of air quality stability, reduces unnecessary equipment adjustments, and reduces system energy consumption and equipment wear. It also ensures timely response when pollution risks arise, achieving a balance between "protecting air quality" and "reducing operating costs."

[0078] In a possible implementation, the real-time dynamic control strategy includes: target fresh air volume, FFU adjustment range, and FFU wind speed adjustment amount.

[0079] The real-time dynamic control strategy is mainly divided into two parts: fresh air volume adjustment and airflow path optimization.

[0080] Among them, the main purpose of fresh air volume adjustment is to reduce the pollution concentration in the clean room by diluting the pollutants. According to the predicted results of the pollutant concentration, the system can dynamically adjust the fresh air volume. In the specific implementation process, based on the predicted results, , location of pollution source and diffusion path simulation , generate real-time dynamic control strategies to adjust the operating status of the fresh air system.

[0081] According to the air quality forecast The pollutant concentration in the air is used to dynamically calculate the fresh air volume. The fresh air volume calculation formula is:

[0082] in, is the target fresh air volume, As the basic fresh air volume; is the adjustment coefficient, which controls the sensitivity of the fresh air volume response; is the target air quality standard value (such as PM concentration threshold); is the predicted pollutant concentration.

[0083] Calculated As the target fresh air volume of the fresh air system, the system adjusts the air volume of the fresh air system in real time to maintain the freshness of the air in the clean room.

[0084] In addition, the purpose of airflow path optimization is to guide the flow direction of the air in the clean room so that pollutants can be quickly discharged and their diffusion in the room can be reduced. Perform path optimization and simulate the results based on the diffusion path , judge the diffusion trend of pollutants. By adjusting the FFU working state and controlling the airflow direction, pollutants can be prevented from spreading to the working area.

[0085] The air duct operating state adjustment process involves controlling the operating states of the FFUs and ambient exhaust vents to optimize the airflow direction. If the contamination source is located near a FFU, the air speed of the FFU in that area can be reduced while increasing the air speed of surrounding FFUs. The ambient exhaust vents are then opened to guide the contaminants toward the ambient exhaust vents to prevent them from spreading to other areas of the cleanroom.

[0086] Optionally, when increasing the wind speed of the surrounding FFUs, the FFUs within a set radius are determined as adjustment objects with the pollution source location as the center, and the FFU wind speed adjustment weight is determined according to the distance between the surrounding FFUs and the pollution source location.

[0087] Set wind speed for different areas , so that the direction of maximum pollutant concentration gradient is consistent with the airflow direction. The airflow path optimization calculation formula is as follows:

[0088] in, For the Wind speed in the area; is the adjustment coefficient, which is used to control the relationship between wind speed and concentration gradient; For the Pollutant concentration gradients across a region.

[0089] The system adjusts the airflow direction in real time based on the calculated FFU switch status and wind speed setting of the air duct to effectively guide the discharge of pollutants.

[0090] After the above fresh air volume adjustment and airflow path optimization process is completed, the control strategy is finally generated. , including the target fresh air volume, FFU adjustment range and FFU wind speed adjustment amount, which are fed back to the reinforcement learning module to further optimize the control effect.

[0091] In this embodiment, the real-time dynamic control strategy includes a target fresh air volume, an FFU adjustment range, and an FFU speed adjustment. Based on this, the fresh air system can be precisely adjusted based on pollution conditions. Adjusting the fresh air volume dilutes pollutant concentrations, while adjusting the FFU adjustment range and speed directs airflow, accelerating pollutant removal. This avoids the inefficiency of traditional control methods that simply increase fresh air volume, improves the targetedness and effectiveness of control measures, and more effectively safeguards cleanroom air quality.

[0092] In a possible implementation, after generating the real-time dynamic control strategy, the following steps are further included: Update and obtain air quality data, energy consumption of fresh air systems and purification equipment; The real-time collected air quality data, the energy consumption of the fresh air system and purification equipment are input into the trained deep reinforcement learning network to obtain the optimal control strategy of the fresh air system; Among them, the deep reinforcement learning network training process is as follows: Design the state space S of deep reinforcement learning, which includes the real-time air quality parameters of the clean room, the control parameters of the real-time dynamic control strategy, and the system energy consumption data; Set the action space A, which includes the action mode a of increasing or decreasing the fresh air volume, adjusting the FFU wind speed, and switching the FFU switch state; Constructing the reward function:

[0093] in, Energy consumption of fresh air system and purification equipment; 、 is the weight coefficient of the reward function, which represents the weight of the energy consumption and air quality parameters of the fresh air system and purification equipment in the reward. is the air quality compliance rate; A deep reinforcement learning framework for the PPO algorithm was built. Based on the historical data in the clean room, S, A, and R were extracted, and the deep reinforcement learning framework was trained to obtain a trained deep reinforcement learning network. The historical data in the clean room included: air quality parameters, energy consumption of the fresh air system and purification equipment, target fresh air volume, FFU adjustment range, and FFU wind speed adjustment.

[0094] In this embodiment, the core idea of ​​reinforcement learning is to gradually optimize the strategy through trial and error, allowing the system to find the optimal control method through continuous interaction. Specifically, the reinforcement learning process includes the steps of state, action, reward function, and strategy update.

[0095] First, a deep reinforcement learning state space S is designed to represent the system's current air quality and equipment operating status. S contains real-time cleanroom air quality parameters, control parameters for the real-time dynamic control strategy, and system energy consumption data. Air quality parameters include PM, TA, TB, VOC concentrations, temperature, and humidity. System control parameters include the current fresh air volume and airflow path. System energy consumption data includes the current energy consumption of the fresh air system and purification equipment.

[0096] Next, set the action space A. Actions are the control adjustments that the system can perform in its current state. Setting action space A indicates that the system can adjust the cleanroom fresh air system. A includes actions such as increasing or decreasing fresh air volume, adjusting FFU air speed, and switching the FFU on and off.

[0097] Then, a reward function is constructed to evaluate the effect of each control action and guide the system to find a balance between air quality and energy consumption. Mainly based on air quality compliance and system energy consumption.

[0098] Specific calculation method: If the air quality meets the standard (i.e., PM, TA, TB, VOC and other indicators are all within the target range), the reward value increases. If the air quality does not meet the standard or the energy consumption is too high, the system will receive a negative reward, which will push the system to reduce energy consumption or improve air quality in the next step.

[0099] In order to ensure the system's strong adaptability in dynamic environments, PPO is selected to optimize the control strategy. PPO has good convergence and stability, making it suitable for this complex continuous control task.

[0100] PPO samples the system's execution data (state, action, reward) under the current policy and uses this data to update the policy, resulting in a higher expected total reward when executing the new policy. The PPO algorithm uses an advantage function to estimate the current policy's performance relative to the baseline policy and imposes limits on policy changes during updates to prevent system instability caused by excessively fast policy updates.

[0101] The objective function of the PPO algorithm is:

[0102] in: Indicates the ratio of the current strategy to the old strategy; is the advantage function, which is used to evaluate the quality of the current action; is a hyperparameter that limits the magnitude of policy updates.

[0103] During each training round, the system performs a series of control actions and updates its strategy based on reward feedback. After multiple rounds of iteration, the system strategy stabilizes and achieves an optimal balance between air quality and energy consumption.

[0104] During the training process of the reinforcement learning algorithm, the system continuously uses real-time feedback data To evaluate and optimize the control strategy. At each time step, the system performs the control action under the current strategy , and record the current status ,action ,award And the new state after execution These interaction data will be used to train reinforcement learning algorithms and improve the system's adaptability in different environments.

[0105] When the system detects that the air quality does not meet the standard or the energy consumption is abnormal, it automatically triggers the adjustment of the control strategy.

[0106] As time goes by, the system updates its strategy through continuously accumulated feedback data, gradually enhancing the adaptability of the control strategy, and ultimately enabling the clean room to achieve the goals of meeting air quality standards and minimizing energy consumption in different environments.

[0107] In this embodiment, by updating air quality data and energy consumption data, real-time feedback is provided for optimizing the control strategy. Then, by designing a state space S that includes air quality, control parameters, and energy consumption, an action space A that covers fresh air volume and FFU adjustment, and a reward function that balances energy consumption and air quality compliance, the PPO algorithm is combined to train a deep reinforcement learning network, enabling the network to learn the optimal control logic based on historical and real-time data. Finally, real-time data is input into the trained network to obtain the optimization strategy, achieving self-learning and self-adaptation of the fresh air system control. This can not only maintain air quality compliance for a long time, but also continuously optimize energy consumption through the guidance of the reward function, avoiding the limitations of traditional control strategies that are difficult to adapt to environmental changes, and improving the efficiency and economy of the system's long-term operation.

[0108] The following are device embodiments of the present invention. For details not fully described therein, reference may be made to the corresponding method embodiments described above.

[0109] Figure 3 The following is a schematic diagram showing the structure of a control device for a clean room fresh air system according to an embodiment of the present invention. For ease of explanation, only the parts related to the embodiment of the present invention are shown, which are described in detail as follows: like Figure 3 As shown, the control device 3 of the clean room fresh air system includes: The acquisition module 301 is used to collect air quality data from different detection points in the clean room in real time and generate a raw data sequence; the air quality data includes: particulate matter concentration, total acidic pollutant TA concentration, total alkaline pollutant TB concentration, VOC concentration, temperature, humidity, wind speed and pressure difference.

[0110] The dimension reduction module 302 is used to extract key feature data from the original data sequence, and use PCA to reduce the dimension of the key feature data to form a reduced dimension feature matrix.

[0111] The fusion prediction module 303 is used to input the reduced dimension feature matrix into ARIMA and LSTM respectively to obtain prediction results, and perform weighted fusion to obtain a fusion prediction result of air quality.

[0112] The fresh air system control module 304 is used to mark the location of potential pollution sources based on the fusion prediction results and detection point locations, and obtain diffusion path simulation results according to the diffusion model. Based on the fusion prediction results, potential pollution source locations and diffusion path simulation results, a real-time dynamic control strategy is generated to control the operating status of the fresh air system.

[0113] In this embodiment, multi-dimensional air quality data of the clean room is first collected in real time to ensure comprehensive monitoring of the air quality status and avoid the risk of pollution missed by a single indicator. Then, the key feature data is reduced in dimension through PCA, which effectively reduces the amount of subsequent model calculations and reduces the system operation load; the reduced-dimensional data is then input into ARIMA and LSTM and weighted fused, combining ARIMA's ability to capture periodic changes with LSTM's adaptability to short-term fluctuations to improve the accuracy of air quality prediction and achieve forward-looking judgment on air quality changes. Finally, a control strategy is generated based on the fusion prediction results, pollution source location, and diffusion path simulation to make the fresh air system control more targeted and avoid the lag of traditional fixed fresh air volume. This can not only respond to pollution risks in a timely manner and ensure the stability of clean room air quality, but also reduce unnecessary energy waste.

[0114] Figure 4 Schematic diagram of an electronic device provided by an embodiment of the present invention. Figure 4 As shown, the electronic device 4 of this embodiment includes a processor 40 and a memory 41. The memory 41 stores a computer program 42. When the processor 40 executes the computer program 42, the steps of the above-described method embodiments are implemented. Alternatively, when the processor 40 executes the computer program 42, the functions of the modules / units in the above-described device embodiments are implemented.

[0115] For example, the computer program 42 may be divided into one or more modules / units, which are stored in the memory 41 and executed by the processor 40 to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program 42 in the electronic device 4.

[0116] The electronic device 4 may include, but is not limited to, a processor 40 and a memory 41. Those skilled in the art will appreciate that Figure 4 It is only an example of the electronic device 4 and does not constitute a limitation on the electronic device 4. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device 4 may also include input and output devices, network access devices, buses, etc.

[0117] The processor 40 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0118] Memory 41 can be an internal storage unit of electronic device 4, such as a hard drive or memory of electronic device 4. Memory 41 can also be an external storage device of electronic device 4, such as a plug-in hard drive, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. equipped on electronic device 4. Furthermore, memory 41 can include both an internal storage unit of electronic device 4 and an external storage device. Memory 41 is used to store computer program 42 and other programs and data required by electronic device 4. Memory 41 can also be used to temporarily store data that has been output or is about to be output.

[0119] For the sake of convenience and brevity, the division of the above functional modules / units is only used as an example. In actual applications, the above functions can be assigned to different functional modules / units as needed. The above modules / units can be implemented in the form of hardware, software, or a combination of hardware and software.

[0120] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods in the above-mentioned method embodiments.

[0121] An embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the methods in the above-mentioned method embodiments.

[0122] The term "computer program" includes computer program code, which may be in source code form, object code form, executable file, or some intermediate form. Computer-readable media may include any entity or device capable of carrying computer program code, recording media, USB flash drives, removable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunications signals, and software distribution media.

[0123] In the above embodiments, the descriptions of each embodiment have their own focus. For parts not described or recorded in detail in one embodiment, please refer to the relevant descriptions of other embodiments. Unless otherwise specified or there is a logical conflict, the terms and / or descriptions between different embodiments are consistent and can be referenced to each other. The technical features of different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0124] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A method for controlling a fresh air system in a clean room, characterized in that: include: The system collects air quality data from different detection points in the clean room in real time and generates a raw data sequence; the air quality data includes: particulate matter concentration, total acidic pollutant TA concentration, total alkaline pollutant TB concentration, volatile organic compound VOC concentration, temperature, humidity, wind speed and pressure difference; Extract key feature data from the original data sequence, and perform dimensionality reduction on the key feature data using principal component analysis (PCA) to form a dimensionality reduction feature matrix; The dimension-reduced feature matrix is ​​input into the autoregressive moving average model (ARIMA) and the long short-term memory network (LSTM) to obtain prediction results, and weighted fusion is performed to obtain a fused prediction result of air quality; Based on the fusion prediction results and the detection point locations, the potential pollution source locations are marked, and the diffusion path simulation results are obtained according to the diffusion model. Based on the fusion prediction results, the potential pollution source locations and the diffusion path simulation results, a real-time dynamic control strategy is generated to control the operating status of the fresh air system; The step of marking the location of a potential pollution source based on the fusion prediction result and the detection point location includes: Determine the pollution source area through spatial gradient analysis based on the fusion prediction results and the detection point locations; The Gaussian mixture model is used to establish the source point characteristic distribution. The pollutant concentration and detection point location of the detection point are used as input to calculate the probability that each detection point location belongs to the source point cluster. When the probability exceeds the preset threshold, the detection point location is marked as a potential pollution source location.

2. The method for controlling the fresh air system in a clean room according to claim 1, characterized in that: Determining the pollution source area through spatial gradient analysis based on the fusion prediction results and the detection point locations includes: The pollutant concentrations at each detection point in the fusion prediction results are interpolated into a continuous distribution, and the gradient and Laplace value are calculated based on the interpolation results to determine the spatial variation trend; The grids with local peaks marked and the changing trend radiating outward are candidate sources; For each candidate source, the wind speed and pressure difference are combined to trace back multiple times along the wind direction and negative gradient direction, and the pollution source area is determined using the change trend threshold, peak threshold and consistency threshold.

3. The method for controlling the fresh air system in a clean room according to claim 1, characterized in that: The diffusion path simulation results are obtained based on the diffusion model, including: Calculate the predicted pollutant concentration at each detection point at different time points based on the diffusion model and construct a predicted pollutant concentration sequence; The diffusion path simulation results are obtained based on the predicted pollutant concentration sequence at each detection point; Wherein, the diffusion model formula is: , Indicates location In time The predicted pollutant concentrations, is the initial pollutant concentration, is the diffusion coefficient, To the location of the pollution source distance.

4. The method for controlling the fresh air system in a clean room according to claim 1, characterized in that: Before marking the location of the potential pollution source based on the fusion prediction result and the detection point location, the method further includes: Determine, based on the fusion prediction results, whether there are indicators among the particulate matter concentration, the total acidic pollutant TA concentration, the total alkaline pollutant TB concentration, and the VOC concentration whose concentration increases beyond a set threshold; Correspondingly, when there is an indicator that the concentration increases beyond the set threshold, the potential pollution source location is marked based on the fusion prediction result and the detection point location and subsequent operations are performed; otherwise, the fresh air system maintains the current state unchanged.

5. The method for controlling the fresh air system in a clean room according to claim 1, characterized in that: The real-time dynamic control strategy includes: target fresh air volume, fan filter unit FFU adjustment range and FFU wind speed adjustment amount.

6. The method for controlling the fresh air system in a clean room according to claim 5, characterized in that: After generating the real-time dynamic control strategy to control the operating state of the fresh air system, the method further includes: Update and obtain air quality data, energy consumption of fresh air systems and purification equipment; The real-time collected air quality data, the energy consumption of the fresh air system and purification equipment are input into the trained deep reinforcement learning network to obtain the optimal control strategy of the fresh air system; The deep reinforcement learning network training process is as follows: Design the state space S of deep reinforcement learning, which includes the real-time air quality parameters of the clean room, the control parameters of the real-time dynamic control strategy, and the system energy consumption data; Set the action space A, which includes the action mode a of increasing or decreasing the fresh air volume, adjusting the FFU wind speed, and switching the FFU switch state; Constructing the reward function: in, Energy consumption of fresh air system and purification equipment; 、 is the weight coefficient of the reward function, which represents the weight of the energy consumption and air quality parameters of the fresh air system and purification equipment in the reward. is the air quality compliance rate; A deep reinforcement learning framework based on the PPO algorithm was built. S, A, and R were extracted based on historical data in the clean room, and the deep reinforcement learning framework was trained to obtain a trained deep reinforcement learning network. The historical data in the clean room included: air quality parameters, energy consumption of the fresh air system and purification equipment, target fresh air volume, FFU adjustment range, and FFU wind speed adjustment.

7. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 6 when the computer program is executed by a processor.

Images