Method, device, medium and computer program for controlling a clean room fresh air system

Abstract

The application provides a clean room fresh air system management and control method, equipment, medium and computer program, and relates to the technical field of clean room ventilation control. The method comprises the following steps: the system collects air quality data of different detection points of the clean room in real time, and generates an original data sequence; key feature data is extracted from the original data sequence, PCA is used to reduce the dimension of the key feature data, and a dimension-reduced feature matrix is formed; the dimension-reduced feature matrix is input into ARIMA and LSTM respectively to obtain prediction results, and the prediction results are weighted and fused to obtain a fusion prediction result of air quality; based on the fusion prediction result and the detection point position, the position of a potential pollution source is marked, and a diffusion path simulation result is obtained according to a diffusion model; based on the fusion prediction result, the position of the potential pollution source and the diffusion path simulation result, a real-time dynamic control strategy is generated to control the operating state of the fresh air system. The application can ensure that the air quality meets the standard while realizing efficient and energy-saving operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of clean room ventilation control, and in particular to a clean room fresh air system management and control method, device, medium and computer program. BACKGROUND

[0002] In semiconductor manufacturing, the air quality of the clean room is of great importance. Tiny particulate matter or chemical pollutants can cause irreversible damage to the processing of chips, directly affecting the yield and performance of products. Therefore, the semiconductor clean room needs to maintain extremely high cleanliness. However, the existing air quality management method mainly relies on a fixed fresh air volume, monitors the air quality by a preset threshold, and triggers control measures when the threshold is exceeded. This approach lacks foresight in air quality changes and often fails to respond to sudden pollution in a timely manner, resulting in substandard air quality in a short period of time, thereby affecting the stability of product manufacturing.

[0003] In addition, the existing system lacks precise positioning and control means for pollution sources. When pollution occurs in the clean room, the system usually handles it by increasing the fresh air volume and filtering frequency, but this approach is inefficient. Due to the complexity of the clean room environment, the distribution of pollution sources may be diverse, and traditional methods are difficult to achieve point control of pollution sources, often causing high load operation of the clean room system, increasing energy consumption and operating costs. These problems make it difficult for the existing air quality management system to ensure that the air quality meets the standards while achieving efficient and energy-saving operation. SUMMARY

[0004] The embodiments of the present application provide a clean room fresh air system management and control method, device, medium and computer program to solve the problem of how to ensure that the air quality meets the standards while achieving efficient and energy-saving operation.

[0005] In a first aspect, the embodiments of the present application provide a clean room fresh air system management and control method, characterized in that it comprises:

[0006] The system collects air quality data of different detection points in the clean room in real time to generate an original data sequence; wherein the air quality data includes particulate matter concentration, total acid pollutant TA concentration, total base pollutant TB concentration, volatile organic compound (VOC) concentration, temperature, humidity, wind speed and pressure difference;

[0007] Key feature data is extracted from the original data sequence, and principal component analysis (PCA) is used to reduce the dimension of the key feature data to form a dimension-reduced feature matrix;

[0008] The dimension reduction feature matrix is respectively input into an autoregressive integrated moving average model (ARIMA) and a long short-term memory network (LSTM) to obtain prediction results, and the prediction results are weighted and fused to obtain a fusion prediction result of air quality.

[0009] Based on the fusion prediction result and the detection point position, a potential pollution source position is marked, a diffusion path simulation result is obtained according to a diffusion model, and a real-time dynamic control strategy is generated based on the fusion prediction result, the potential pollution source position and the diffusion path simulation result, to adjust the operation state of the fresh air system.

[0010] In a possible implementation, the marking of the potential pollution source position based on the fusion prediction result and the detection point position comprises:

[0011] According to the fusion prediction result and the detection point position, a pollution source area is determined through spatial gradient analysis;

[0012] A Gaussian mixture model is used to establish a source point feature distribution, and the detection point pollutant concentration and the detection point position are used as inputs to calculate the probability of each detection point position belonging to a source point cluster. When the probability is higher than a preset threshold, the detection point position is marked as a potential pollution source position.

[0013] In a possible implementation, the determination of the pollution source area according to the fusion prediction result and the detection point position through spatial gradient analysis comprises:

[0014] The pollutant concentrations of the detection points in the fusion prediction result are interpolated into a continuous distribution, and the gradient and the Laplacian value are calculated according to the interpolation result to determine the spatial variation trend;

[0015] A grid with a local peak value and a variation trend radiating outward is marked as a candidate source;

[0016] For each candidate source, the wind speed and the pressure difference are combined to backtrack multiple time points along the wind direction and the negative gradient direction, and the variation trend threshold, the peak value threshold and the consistency threshold are used to determine a pollution source area.

[0017] In a possible implementation, the diffusion path simulation result is obtained according to a diffusion model, comprising:

[0018] The predicted pollutant concentrations of the detection points at different time points are calculated according to the diffusion model to construct a predicted pollutant concentration sequence;

[0019] The diffusion path simulation result is obtained according to the predicted pollutant concentration sequence of each detection point;

[0020] The diffusion model formula is:

[0021]

[0022] , representing a position at a time predicted pollutant concentration, is an initial pollutant concentration, is a diffusion coefficient, is a distance from a pollution source to the position .

[0023] In a possible implementation, before the step of marking a potential pollution source position based on the fusion prediction result and the detection point position, the method further comprises:

[0024] determining whether there is an index of concentration increase exceeding a set threshold value in the concentration of particulate matter, the concentration of total acidic pollutants TA, the concentration of total basic pollutants TB, and the concentration of VOC according to the fusion prediction result;

[0025] Correspondingly, when there is an index of concentration increase exceeding a set threshold value, the step of marking a potential pollution source position based on the fusion prediction result and the detection point position and the subsequent operation are performed; otherwise, the current state of the fresh air system remains unchanged.

[0026] In a possible implementation, the real-time dynamic control strategy comprises a target fresh air volume, a Fan Filter Unit (FFU) adjustment range, and an FFU air speed adjustment amount.

[0027] In a possible implementation, after the step of generating the real-time dynamic control strategy, the method further comprises:

[0028] updating the acquired air quality data and energy consumption of the fresh air system and the purification equipment;

[0029] inputting the real-time collected air quality data and energy consumption of the fresh air system and the purification equipment into the trained deep reinforcement learning network to obtain an optimized control strategy of the fresh air system;

[0030] wherein the deep reinforcement learning network training process is as follows:

[0031] designing a state space S of deep reinforcement learning, wherein the state space S comprises real-time air quality parameters of the clean room, control parameters of the real-time dynamic control strategy, and system energy consumption data;

[0032] setting an action space A, wherein the action space A comprises action modes a of fresh air volume increase and decrease, FFU air speed adjustment, and FFU switch state switching;

[0033] constructing a reward function:

[0034]

[0035] wherein, is the energy consumption of the fresh air system and the purification equipment; 、 is the weight coefficient of the reward function, indicating the weight of the energy consumption of the fresh air system and the purification equipment and the air quality parameter in the reward, is the air quality compliance rate;

[0036] A deep reinforcement learning framework based on a proximal policy optimization (PPO) algorithm is built, S, A and R are extracted according to historical data in the clean room, the deep reinforcement learning framework is trained, and a trained deep reinforcement learning network is obtained; wherein the historical data in the clean room includes air quality parameters, energy consumption of the fresh air system and the purification equipment, target fresh air volume, FFU adjustment range and FFU wind speed adjustment amount.

[0037] In a second aspect, an embodiment of the present application provides a clean room fresh air system management and control device, comprising:

[0038] The acquisition module is configured to acquire air quality data of different detection points in the clean room in real time to generate an original data sequence; wherein 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.

[0039] The dimension reduction module is configured to extract key feature data from the original data sequence, perform dimension reduction on the key feature data using PCA, and form a dimension reduction feature matrix.

[0040] The fusion prediction module is configured to input the dimension reduction feature matrix into ARIMA and LSTM respectively to obtain prediction results, and perform weighted fusion to obtain a fusion prediction result of the air quality.

[0041] The fresh air system control module is configured to mark a potential pollution source position based on the fusion prediction result and the detection point position, obtain a diffusion path simulation result based on a diffusion model, and generate a real-time dynamic control strategy based on the fusion prediction result, the potential pollution source position and the diffusion path simulation result to control the operation state of the fresh air system.

[0042] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the method in the first aspect or any possible implementation manner of the first aspect when executing the computer program.

[0043] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium storing a computer program, the computer program being executed by a processor to implement the method in the first aspect or any possible implementation manner of the first aspect.

[0044] In a fifth aspect, an embodiment of the present application provides a computer program product comprising a computer program, the computer program being executed by a processor to implement the method in the first aspect or any possible implementation manner of the first aspect.

[0045] In the embodiment of the present application, the multi-dimensional air quality data of the clean room is first collected in real time to ensure comprehensive monitoring of the air quality state and avoid missing pollution risks due to a single index. Then, the PCA is used to reduce the dimension of the key feature data, effectively reducing the subsequent model calculation amount and reducing the system operation load. Subsequently, the reduced dimension data is input into the ARIMA and LSTM and weighted fusion, combining the capture ability of the ARIMA for periodic changes and the adaptability of the LSTM for short-term fluctuations to improve the air quality prediction accuracy and realize the forward-looking judgment of the air quality change. Finally, based on the fusion prediction result, the pollution source positioning and the diffusion path simulation, the control strategy is generated, so that the fresh air system control is more targeted, avoiding the lag of the traditional fixed fresh air volume, which can not only timely respond to pollution risks and ensure the stability of the clean room air quality, but also reduce unnecessary energy waste. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 is the application scenario diagram of the clean room fresh air system management and control method provided by the embodiment of the present application;

[0047] Figure 2 is the implementation flowchart of the clean room fresh air system management and control method provided by the embodiment of the present application;

[0048] Figure 3 is a structural schematic diagram of the clean room fresh air system management and control device provided by the embodiment of the present application;

[0049] Figure 4 is a schematic diagram of an electronic device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0050] The present application is applicable to multiple fields requiring high-cleanliness air environment, especially the industries such as semiconductor manufacturing, biomedicine, precision instrument production and aerospace, which are extremely sensitive to micro pollutants. In these fields, the stability and cleanliness of the clean room air quality directly affect the product quality and production efficiency, and the present application can provide an intelligent and efficient air quality management scheme to effectively improve the production yield and reduce the operating cost.

[0051] The embodiments of the present application will be described in detail below with reference to the drawings.

[0052] Figure 1 The application scenario diagram of the clean room fresh air system management method provided by the embodiment of the application is shown in FIG. 1. Figure 1 As shown in the figure, a plurality of concentration sensors, FFUs, fresh air equipment and environmental exhaust ports are distributed in the clean room. The FFUs, fresh air equipment and environmental exhaust ports constitute a complete fresh air system.

[0053] 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 concentration prediction result of the pollutants, the system can dynamically adjust the fresh air volume.

[0054] In the actual implementation process, the air duct working state adjustment is involved, mainly including: controlling the working state of the FFU and the environmental exhaust port to optimize the air flow direction. If the pollution source position is close to a certain FFU, the FFU in this area can reduce the wind speed, while increasing the wind speed of the surrounding FFU, and opening the environmental exhaust port to guide the pollutants to move towards the environmental exhaust port direction, so as to prevent the pollutants from spreading to other areas of the clean room.

[0055] The control terminal is used to remotely collect the concentration detection data of the concentration sensor, the operating parameters and energy consumption data of the FFU and the fresh air equipment, so as to realize efficient and energy-saving operation while ensuring that the air quality meets the standard.

[0056] The sensors in the application include particulate matter concentration, total acidic pollutant TA concentration, total basic pollutant TB concentration, VOC concentration, temperature, humidity, wind speed and differential pressure sensor.

[0057] In order to comprehensively monitor the air quality state in the clean room, the scheme needs to collect multiple air quality indexes. The particulate matter concentration PM mainly includes PM0.5, PM0.3, PM0.2 and PM0.1, which reflects the concentration of particulate matter with a diameter less than 0.5 / 0.3 / 0.2 / 0.1 microns in the air. The TA concentration is the total acidic pollutant concentration in the clean room air, mainly including hydrogen fluoride (HF), hydrogen chloride (HCl), sulfur dioxide (SO2) and the like. The TB concentration is the total basic pollutant concentration in the clean room air, mainly including ammonia (NH3), organic amine (such as triethylamine, dimethyl ethanolamine), alkyl amine compounds and the like. The VOC concentration mainly identifies potential chemical pollutants in the air.

[0058] In addition, the temperature and humidity are detected to help judge the running state of the air conditioning system, and at the same time have a certain correlation with pollution diffusion. The wind speed and pressure difference are detected to reflect the air flow and pressure difference, and provide the air flow distribution information for the control system.

[0059] The sensors are arranged in different areas of the clean room to ensure data coverage and high-precision acquisition in key areas.

[0060] The sensors are set at the fresh air inlet and outlet to monitor the air quality entering and leaving the clean room, capturing the degree of air pollution input and output. Sensors are set in key internal areas, such as areas with high human traffic or equipment concentration, to monitor local air quality fluctuations. Sensors are distributed in the four corners and central area of the clean room to monitor air quality uniformity and pollution diffusion.

[0061] In the implementation process, these sensors are connected through an Internet of Things (IoT) module, transmitting data in real time to edge computing nodes to support data processing and subsequent analysis.

[0062] To ensure real-time data collection, the system defaults to collecting data every 10 seconds, which is the basic sampling frequency in a stable state, sufficient to provide real-time data support. In addition, the system monitors air quality fluctuations in real time and automatically adjusts the sampling frequency based on the fluctuation amplitude. For example: when detecting that the air quality indicators (such as PM, TA, TB) fluctuate greatly, the sampling frequency is automatically increased to every 2 seconds to respond to sudden pollution events and ensure that the diffusion trend of pollutants in the clean room is captured in a timely manner.

[0063] Dynamic sampling frequency formula:

[0064]

[0065] Where, is the current sampling frequency, is the initial sampling frequency (10 seconds), is the adjustment coefficient, is the air quality change rate (such as the change rate of PM). When is large, the system automatically increases the sampling frequency.

[0066] Before executing the clean room fresh air system control method provided in the present application, the sensor detection data needs to be preprocessed. Optionally, the data collected by the sensor is transmitted to the edge computing node through the Internet of Things module for data preprocessing to improve data processing efficiency. Preprocessing includes missing value processing, outlier removal, and noise processing.

[0067] In the missing value processing process, linear interpolation is used to fill in short-term missing data, as follows:

[0068]

[0069] In the outlier removal process, the three-sigma method is used to remove outliers to ensure data reliability:

[0070] is marked as an outlier.

[0071] In the noise processing process, the moving average method is used to smooth the data and reduce the short-term random fluctuations that the sensor can generate. The moving average formula is:

[0072]

[0073] wherein, is the smoothed data, is the window size, generally 3 to 5, is the original data point in the window.

[0074] The clean room fresh air system management and control method provided by the embodiments of the present application is based on the above. In order to make the specific process of the clean room fresh air system management and control method clearer, the method embodiments are described in detail below with reference to the accompanying drawings.

[0075] Referring to Figure 2 , it shows the implementation flowchart of the clean room fresh air system management and control method provided by the embodiments of the present application, which includes the following steps:

[0076] S201, the system collects air quality data of different detection points in the clean room in real time to generate an original data sequence; wherein 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.

[0077] The execution subject of each embodiment of the present application can be a server, a processor, a microprocessor or other devices with data processing function. In the actual implementation process, the specific implementation mode of the execution subject can be selected according to actual needs, and the present embodiment does not make special limitation, as long as it is a device with data processing function. In order to facilitate the understanding of the embodiments of the present application, the control terminal shown in Figure 1 is taken as the execution subject for description.

[0078] In the specific implementation process, the specific classification of air quality data is described in the foregoing Figure 1 . It is suitable for specific application fields. According to the different compositions of gases and particulate matters in the clean room environment, the detection and classification of particulate matter concentration, total acidic pollutant TA concentration, total alkaline pollutant TB concentration and VOC concentration are different.

[0079] In order to improve the adjustment efficiency of the clean room fresh air system, the present application implements the corresponding indoor environment change. After the control terminal collects the air quality data of different detection points in the clean room, the original data sequence is generated as , and is taken as the input of feature extraction and dimension reduction.

[0080] S202, extract key feature data from the original data sequence, use PCA to reduce the dimension of the key feature data, and form a reduced dimension feature matrix.

[0081] From the original data , extract key features, aiming to simplify the data dimension and highlight the core parameters affecting air quality changes.

[0082] The extracted key features are closely related to air quality changes, such as PM, TA, TB change rate, temperature and humidity fluctuation range, etc. After extracting the features, we get data . It contains key air quality parameters, providing concise input for subsequent prediction.

[0083] In order to facilitate analysis and processing, PCA is used to reduce the dimension of feature data , forming the final reduced dimension feature matrix . PCA formula:

[0084]

[0085] where, is the feature vector matrix, is the transpose matrix of , and is the reduced dimension feature matrix, representing the simplified core air quality parameters.

[0086] The reduced dimension data contains the main air quality information, reduces the calculation complexity, and reduces data redundancy. At this time, the prediction model only needs to process a small number of core features, which can more efficiently predict air quality trends and improve model accuracy.

[0087] S203, input the reduced dimension feature matrix into ARIMA and LSTM to get the prediction results, and perform weighted fusion to get the fusion prediction result of air quality.

[0088] Input the reduced dimension feature matrix into ARIMA and LSTM, aiming to use the reduced dimension data to predict the future air quality trend, providing basis for early adjustment of control strategy. ARIMA model captures the periodic changes of air quality (such as diurnal cycle), providing a basis for short-term prediction. LSTM is mainly used to capture short-term fluctuations in air quality, improving prediction accuracy.

[0089] where, ARIMA formula:

[0090]

[0091] where, is the predicted value of the current air quality; For the predicted value of air quality at time t; is the autoregressive coefficient, responsible for capturing the trend of the data; is the coefficient of the moving average (MA) part, used to handle noise; is the current white noise term, i.e., random error; For the white noise term at time t.

[0092] The LSTM calculation steps include:

[0093] Forget gate:

[0094] Input gate:

[0095] State update:

[0096] Output gate:

[0097] where, is the forget gate output vector; is the sigmoid activation function; is the forget gate weight matrix; is the LSTM hidden state vector at the previous time; is the reduced dimension feature matrix; is the forget gate bias term; is the input gate output vector; is the input gate bias term; is the current cell state vector at time t; is the cell state vector at the previous time; is the candidate cell state vector; is the hyperbolic tangent activation function; is the hidden state vector at the current time; is the output gate output vector.

[0098] Finally, the outputs of ARIMA and LSTM are weighted and fused to obtain the final prediction result .

[0099] The fusion formula is:

[0100]

[0101] where, and are the prediction outputs of ARIMA and LSTM, and are the weight coefficients.

[0102] S204, based on the fusion prediction result and the detection point position, marking the potential pollution source position, and obtaining the diffusion path simulation result according to the diffusion model, 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.

[0103] In the specific implementation process, based on the prediction result The air quality change characteristics are combined with the sensor position data (corresponding to the detection point position) to mark the potential pollution source position Based on the prediction result , the pollution source position And the diffusion path simulation , generate a real-time dynamic control strategy to adjust the operation state of the fresh air system. The control system automatically increases the fresh air volume, dilutes the pollutants, and adjusts the FFU state to adjust the pollutant diffusion path, while taking into account the overall energy consumption of the fresh air system.

[0104] In this embodiment, the multi-dimensional air quality data of the clean room is first collected in real time to ensure comprehensive monitoring of the air quality state and avoid missing pollution risks due to single indicators. Then, PCA is used to reduce the dimensionality of key feature data, effectively reducing the subsequent model calculation amount and reducing the system operation load; subsequently, the reduced dimension data is input into ARIMA and LSTM and weighted fusion, combining the periodic change capturing ability of ARIMA and the adaptability of LSTM to short-term fluctuations to improve the air quality prediction accuracy and realize the forward-looking judgment of air quality changes. Finally, based on the fusion prediction result, the pollution source positioning and the diffusion path simulation, a control strategy is generated to make the fresh air system control more targeted, avoiding the lag of traditional fixed fresh air volume, which can not only timely respond to pollution risks and ensure the stability of clean room air quality, but also reduce unnecessary energy waste

[0105] In one possible implementation, based on the fusion prediction result and the detection point position, the potential pollution source position is marked, comprising:

[0106] According to the fusion prediction result and the detection point position, the pollution source area is determined by spatial gradient analysis;

[0107] A Gaussian mixture model is used to establish the source point feature distribution, taking the detection point pollutant concentration and the detection point position as input, calculating the probability of each detection point position belonging to the source point cluster, and when the probability is greater than a preset threshold, marking the detection point position as the potential pollution source position.

[0108] In this embodiment, the spatial gradient analysis is first used to preliminarily determine the pollution source area by combining the fusion prediction result and the detection point position, to provide a spatial range basis for the potential pollution source positioning. Then, the Gaussian mixture model is used to construct the source point feature distribution by taking the pollutant concentration and position of the detection point as the input, to mark the potential pollution source by calculating the probability of each position belonging to the source point cluster and combining the preset threshold, to upgrade the pollution source positioning from "area judgment" to "probability quantitative judgment", to effectively reduce the misjudgment risk caused by single spatial analysis, to make the potential pollution source position marking more accurate and reliable, and to provide a more explicit target direction for the generation of the targeted control strategy.

[0109] The above embodiment generally describes that, in the process of marking the potential pollution source position based on the fusion prediction result and the detection point position, two steps of determining the pollution source area and marking the pollution source position are included. The following will be described in detail with respect to the determination of the pollution source area by the spatial gradient analysis.

[0110] In a possible implementation, the pollution source area is determined by the spatial gradient analysis according to the fusion prediction result and the detection point position, including:

[0111] The pollutant concentration of each detection point in the fusion prediction result is interpolated as a continuous distribution, and the gradient and Laplace value are calculated according to the interpolation result to determine the spatial variation trend;

[0112] The grid with a local peak value and a variation trend radiating outward is marked as a candidate source;

[0113] For each candidate source, the wind speed and pressure difference are traced back to multiple time points along the wind direction and the negative gradient direction, and the pollution source area is determined by the variation trend threshold, the peak value threshold and the consistency threshold.

[0114] In the actual implementation process, the fusion prediction result at least includes the pollutant concentration of the detection point, the detection point coordinates, the wind speed / wind direction and the pressure difference at the same time; optionally, it also includes the temperature, humidity and other environmental quantities.

[0115] Before the Gaussian mixture model is used to determine the potential pollution source position, the pollution source area to which the potential pollution source position belongs is first determined.

[0116] In this embodiment, by refining the spatial gradient analysis process, the discrete pollutant concentration of the detection point is first interpolated as a continuous distribution, to fill the data blind area of the area without the detection point, to ensure the complete perception of the overall concentration field of the clean room; then, the spatial variation trend is determined by calculating the gradient and Laplace value, to accurately identify the grid with a local peak value and a variation trend radiating outward as a candidate source, to preliminarily screen out the area with pollution diffusion characteristics; finally, the wind speed and pressure difference are traced back to multiple time points along the specific direction, and the interference factors such as temporary pollution residues are effectively excluded by the multi-threshold comprehensive judgment, to further improve the accuracy and accuracy of the pollution source area positioning, and to improve the accuracy of the control strategy.

[0117] In a possible implementation, the diffusion path simulation result is obtained according to the diffusion model, including:

[0118] The predicted pollutant concentration of each detection point at different time points is calculated according to the diffusion model, and a predicted pollutant concentration sequence is constructed;

[0119] The diffusion path simulation result is obtained according to the predicted pollutant concentration sequence of each detection point;

[0120] The diffusion model formula is:

[0121]

[0122] , denotes the position at time The predicted pollutant concentration is , the initial pollutant concentration is , the diffusion coefficient is , and the distance from the pollution source to the position .

[0123] In this embodiment, the predicted pollutant concentration of each detection point at different time points is calculated by the diffusion model, and a concentration sequence is constructed. By using the correlation between the initial concentration, the diffusion coefficient and the distance in the model formula, the change of the pollution concentration at different positions at different times is quantitatively predicted, and then the pollution diffusion path and range are clearly restored. Compared with the traditional method of relying only on threshold monitoring, the pollution diffusion trend can be grasped in advance, so that the fresh air system control can be arranged in advance according to the diffusion path, the pollution can be prevented from spreading to the key area, and the initiative and effectiveness of the clean room air quality control are further improved.

[0124] In a possible implementation, before the potential pollution source position is marked based on the fusion prediction result and the detection point position, the method further includes:

[0125] According to the fusion prediction result, it is determined whether there is an index with a concentration increase exceeding a set threshold among the particulate matter concentration, the total acid pollutant TA concentration, the total alkali pollutant TB concentration and the VOC concentration;

[0126] Correspondingly, when there is an index with a concentration increase exceeding a set threshold, the operation of marking the potential pollution source position based on the fusion prediction result and the detection point position is performed; otherwise, the fresh air system remains in the current state.

[0127] The main purpose of the fresh air system adjustment is to dilute the pollutants by adjusting the fresh air volume to reduce the pollution concentration in the clean room. According to the concentration prediction result of the pollutants, the system can dynamically adjust the operation state of the fresh air system. The operation state adjustment rule of the fresh air system is:

[0128] If the prediction result If the system indicates that air quality will worsen in the future (e.g., increased PM or VOC concentrations), it automatically increases the fresh air intake to dilute pollutants. If the air quality forecast is relatively stable, the system maintains a basic fresh air intake to reduce energy consumption.

[0129] In this embodiment, the method first determines whether the core pollution indicators have exceeded the threshold before performing pollution source location. Subsequent location and control operations are only initiated when there is a pollution risk. If the indicators are stable, the current state of the fresh air system is maintained. This avoids the problem of traditional systems continuously executing complex control procedures regardless of air quality stability, reduces unnecessary equipment adjustment frequency, lowers system energy consumption and equipment wear, and ensures timely response when pollution risks occur, achieving a balance between "ensuring air quality" and "reducing operating costs".

[0130] In one possible implementation, the real-time dynamic control strategy includes: target fresh air volume, FFU adjustment range, and FFU fan speed adjustment amount.

[0131] The real-time dynamic control strategy is mainly divided into two parts: fresh air volume adjustment and airflow path optimization.

[0132] The primary purpose of fresh air volume regulation is to reduce the concentration of pollutants in the cleanroom by diluting them. Based on the predicted pollutant concentration, the system can dynamically adjust the fresh air volume. In practice, this adjustment is based on the predicted results. Location of pollution sources and diffusion path simulation This generates real-time dynamic control strategies to adjust the operating status of the fresh air system.

[0133] According to air quality forecasts The concentration of pollutants in the air is used to dynamically calculate the fresh air volume. The formula for calculating the fresh air volume is:

[0134]

[0135] in, To achieve the target fresh air volume, Basic fresh air volume; The adjustment coefficient controls the sensitivity of the fresh air volume response; The target air quality standard value (such as PM concentration threshold); This represents the predicted pollutant concentration.

[0136] 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.

[0137] In addition, the purpose of airflow path optimization is to enable the pollutants to be quickly discharged and reduce the diffusion in the room by guiding the flow direction of the clean room air. After determining the location of the pollution source, path optimization is performed based on the location of the pollution source , the diffusion trend of the pollutants is determined according to the diffusion path simulation result . By adjusting the working state of the FFU, the airflow direction is controlled to avoid the diffusion of the pollutants to the working area.

[0138] The air duct working state adjustment process includes: controlling the working state of the FFU and the environmental exhaust port to optimize the airflow direction. If the pollution source location is close to a certain FFU, the FFU in this area can reduce the wind speed, while increasing the wind speed of the surrounding FFUs, and opening the environmental exhaust port to guide the pollutants to move towards the environmental exhaust port direction to prevent the pollutants from diffusing to other areas of the clean room.

[0139] Optionally, when increasing the wind speed of the surrounding FFUs, the FFUs within a set radius range centered on the pollution source location are determined as the adjustment objects, and the FFU wind speed adjustment weight is determined according to the distance between the surrounding FFUs and the pollution source location.

[0140] The wind speed is set for different areas so that the maximum direction of the pollutant concentration gradient is consistent with the airflow direction, and the airflow path optimization calculation formula is as follows:

[0141]

[0142] wherein, is the wind speed of the i-th area; is an adjustment coefficient for controlling the relationship between the wind speed and the concentration gradient; is the pollutant concentration gradient of the i-th area. The system adjusts the airflow direction in real time according to the calculated air duct corresponding FFU switch state and wind speed setting to effectively guide the discharge of the pollutants.

[0143] After the above new air volume adjustment and airflow path optimization process is executed, the final control strategy is generated, including the target new air volume, the FFU adjustment range and the FFU wind speed adjustment amount, which is fed back to the reinforcement learning module to further optimize the control effect.

[0144]

[0145] ​​In the embodiment, the real-time dynamic control strategy includes a target fresh air volume, an FFU adjustment range, and an FFU air speed adjustment amount. Based on this, the fresh air system can be accurately adjusted according to the pollution situation. The pollution concentration is diluted by adjusting the fresh air volume, and the air flow direction is guided by adjusting the FFU adjustment range and the air speed to accelerate the discharge of pollutants and avoid the inefficient problem of "generally increasing the fresh air volume" in traditional control, thereby improving the pertinence and effectiveness of the control measures and more efficiently ensuring the air quality in the clean room.

[0146] In a possible implementation, after the real-time dynamic control strategy is generated, the method further includes:

[0147] updating the energy consumption of the air quality data acquisition, the fresh air system, and the purification equipment;

[0148] inputting the real-time collected air quality data, the energy consumption of the fresh air system and the purification equipment into the trained deep reinforcement learning network to obtain an optimized control strategy of the fresh air system;

[0149] The deep reinforcement learning network training process is as follows:

[0150] A state space S of the deep reinforcement learning is designed, S includes real-time air quality parameters of the clean room, control parameters of the real-time dynamic control strategy, and system energy consumption data;

[0151] An action space A is set, A includes action modes a of fresh air volume increase and decrease, FFU air speed adjustment, and FFU switch state switching;

[0152] A reward function is constructed:

[0153]

[0154] wherein, is the energy consumption of the fresh air system and the purification equipment; 、 is a weight coefficient of the reward function, indicating the weights of the energy consumption of the fresh air system and the purification equipment and the air quality parameters in the reward, is an air quality compliance rate;

[0155] A deep reinforcement learning framework of the PPO algorithm is built, S, A, and R are extracted according to historical data in the clean room, the deep reinforcement learning framework is trained, and a trained deep reinforcement learning network is obtained; wherein the historical data in the clean room includes air quality parameters, energy consumption of the fresh air system and the purification equipment, a target fresh air volume, an FFU adjustment range, and an FFU air speed adjustment amount.

[0156] In the embodiment, the core idea of reinforcement learning is to gradually optimize the strategy through "trial and error" to find the optimal control method in continuous interaction. Specifically, the reinforcement learning process includes steps such as state, action, reward function, and strategy update.

[0157] First, design the state space S of deep reinforcement learning, representing the current air quality and equipment operating status of the system. S contains real-time air quality parameters of the clean room, control parameters of real-time dynamic control strategy, and system energy consumption data. Among them, the air quality parameters include PM, TA, TB, VOC concentration, temperature, humidity, etc. The system control parameters include the current fresh air volume, air flow path, etc. The system energy consumption data includes the energy consumption of the current fresh air system and purification equipment.

[0158] Second, set the action space A, which is the control adjustment that the system can perform in the current state. Set the action space A, which represents that the system can adjust the clean room fresh air system. A includes the action mode a of fresh air volume increase and decrease, FFU air speed adjustment, and FFU switch state switching.

[0159] Then, build the reward function to evaluate the effect of each control action and guide the system to find a balance between air quality and energy consumption. The reward function is mainly based on the air quality compliance degree and system energy consumption.

[0160] The specific calculation method is:

[0161] If the air quality meets the standard (i.e. PM, TA, TB, VOC, etc. indicators are 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 negative rewards, prompting the system to reduce energy consumption or improve air quality in the next step.

[0162] In order to ensure that the system is adaptable in dynamic environment, PPO is selected to optimize the control strategy. PPO has good convergence and stability, and is suitable for such complex continuous control tasks.

[0163] PPO samples the execution data (state, action, reward) of the system under the current policy, uses these data to update the policy, so that the total reward expectation when executing the new policy is higher. PPO algorithm uses advantage function to estimate the pros and cons of the current policy relative to the baseline policy, and limits the change of the policy during updating, avoiding the instability of the system caused by too fast policy update.

[0164] The objective function of PPO algorithm is:

[0165]

[0166] Where: represents the ratio of the current policy and the old policy; is the advantage function, which is used to evaluate the good and bad of the current action; is a hyperparameter that limits the amplitude of policy update.

[0167] In each round of training, the system performs a series of control actions, updates the strategy according to the reward feedback. After multiple iterations, the system strategy tends to be stable, and can achieve the optimal balance between air quality and energy consumption.

[0168] 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 a control action under the current strategy , records the current state , action , reward and new state after execution . These interaction data will be used for training of the reinforcement learning algorithm to improve the adaptability of the system in different environments.

[0169] When the system detects that the air quality is not up to standard or the energy consumption is abnormal, it automatically triggers the adjustment of the control strategy.

[0170] Over time, the system updates the strategy through the accumulation of feedback data, gradually enhances the adaptability of the control strategy, and ultimately enables the clean room to achieve the goal of air quality up to standard and energy consumption minimization in different environments.

[0171] In this embodiment, by updating the air quality data and energy consumption data, real-time feedback basis is provided for optimizing the control strategy. Then by designing a state space S containing air quality, control parameters and energy consumption, an action space A covering fresh air volume and FFU adjustment, and a reward function balancing energy consumption and air quality up to standard, combined with PPO algorithm training deep reinforcement learning network, the network can 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 optimized strategy, realizing self-learning and self-adaptation of the fresh air system control, which can not only maintain air quality up to standard for a long time, but also continuously optimize energy consumption guided by 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 long-term operation of the system.

[0172] The following is an apparatus embodiment of the present application. For details not described in detail, reference can be made to the corresponding method embodiments described above.

[0173] Figure 3 The structure of the control device of the clean room fresh air system provided by the embodiment of the present application is shown. For the sake of convenience, only the part related to the embodiment of the present application is shown, and the details are as follows:

[0174] As shown in Figure 3 , the control device 3 of the clean room fresh air system comprises:

[0175] The collection module 301 is configured to collect air quality data of different detection points in the clean room in real time to generate an original data sequence, wherein the air quality data comprises particulate matter concentration, total acidic pollutant TA concentration, total alkaline pollutant TB concentration, VOC concentration, temperature, humidity, wind speed and pressure difference.

[0176] The dimension reduction module 302 is configured to extract key feature data from the original data sequence, perform dimension reduction on the key feature data by using PCA, and form a dimension-reduced feature matrix.

[0177] The fusion prediction module 303 is configured to input the dimension-reduced feature matrix into ARIMA and LSTM respectively to obtain prediction results, and perform weighted fusion to obtain a fusion prediction result of the air quality.

[0178] The fresh air system control module 304 is configured to mark a potential pollution source position based on the fusion prediction result and the detection point position, obtain a diffusion path simulation result according to a diffusion model, and generate a real-time dynamic control strategy to control the operation state of the fresh air system based on the fusion prediction result, the potential pollution source position and the diffusion path simulation result.

[0179] In this embodiment, the multi-dimensional air quality data of the clean room is collected in real time to ensure comprehensive monitoring of the air quality state and avoid missing pollution risks due to a single index. Then, the PCA is used to reduce the dimension of the key feature data, effectively reducing the subsequent model calculation amount and reducing the system operation load. Subsequently, the dimension-reduced data is input into ARIMA and LSTM and weighted fusion is performed, combining the periodic change capturing capability of ARIMA and the adaptability of LSTM to short-term fluctuations to improve the air quality prediction accuracy and realize forward-looking judgment of the air quality change. Finally, the control strategy is generated based on the fusion prediction result, the pollution source positioning and the diffusion path simulation, so that the fresh air system control is more targeted, avoiding the lag of the traditional fixed fresh air volume, which can not only timely respond to pollution risks and ensure the stability of the air quality in the clean room, but also reduce unnecessary energy waste.

[0180] Figure 4 is a schematic diagram of an electronic device provided by an embodiment of the present application. As shown in Figure 4 the electronic device 4 of this embodiment includes a processor 40 and a memory 41. The memory 41 stores a computer program 42. The processor 40 implements the steps in each of the method embodiments described above when executing the computer program 42. Alternatively, the processor 40 implements the functions of each module / unit in each of the device embodiments described above when executing the computer program 42.

[0181] For example, the computer program 42 can be divided into one or more modules / units, which are stored in the memory 41 and executed by the processor 40 to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program 42 in the electronic device 4.

[0182] The electronic device 4 can include, but is not limited to, the processor 40 and the memory 41. Those skilled in the art can understand that the electronic device 4 can include more or fewer components than those shown, or combine some components, or include different components, for example, the electronic device 4 can also include an input / output device, a network access device, a bus, etc. Figure 4 The electronic device 4 is only an example and does not constitute a limitation on the electronic device 4, and can include more or fewer components than those shown, or combine some components, or include different components, for example, the electronic device 4 can also include an input / output device, a network access device, a bus, etc.

[0183] The processor 40 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0184] The memory 41 can be an internal storage unit of the electronic device 4, such as a hard disk or a memory of the electronic device 4. The memory 41 can also be an external storage device of the electronic device 4, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 4. Further, the memory 41 can include both the internal storage unit and the external storage device of the electronic device 4. The memory 41 is used to store the computer program 42 and other programs and data required by the electronic device 4. The memory 41 can also be used to temporarily store data that has been output or will be output.

[0185] For the convenience and brevity of description, only the above-mentioned division of the functional modules / units is exemplified, and in actual application, the above-mentioned functions can be completed by different functional modules / units according to needs. The above-mentioned modules / units can be realized in the form of hardware, software or a combination of hardware and software.

[0186] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the method in each method embodiment.

[0187] The embodiment of the present application further provides a computer program product, which comprises a computer program. The computer program is executed by a processor to implement the method in each method embodiment.

[0188] The computer program comprises computer program code, which can be in the form of source code, object code, executable files or some intermediate forms. The computer readable medium can comprise any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, Read-Only Memory (ROM), Random Access Memory (RAM), electric carrier wave signal, telecommunication signal and software distribution medium, etc.

[0189] In the above embodiments, the description of each embodiment has its own focus, and the parts not described or recorded in a certain embodiment can be referred to the relevant description of other embodiments. If there is no special description and no logical conflict, the terms and / or descriptions of different embodiments are consistent and can be mutually referred to, and the technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationship.

[0190] The above embodiments are only used to illustrate the technical solutions of the present application, rather than limit the same. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalent ones. The modification or replacement does not change the essence of the corresponding technical solutions, and should be included in the protection scope of the present application.

Claims

1. A method for controlling a fresh air system in a cleanroom, characterized in that, The system collects air quality data of different detection points in the clean room in real time to generate an original data sequence, wherein the air quality data includes particle concentration, total acidic pollutant TA concentration, total basic pollutant TB concentration, volatile organic compound VOC concentration, temperature, humidity, wind speed and pressure difference; Key feature data is extracted from the original data sequence, and principal component analysis PCA is used to reduce the dimension of the key feature data to form a dimension-reduced feature matrix; The dimension-reduced feature matrix is input into an autoregressive integrated moving average model ARIMA and a long short-term memory network LSTM to obtain a prediction result, and the prediction results are weighted and fused to obtain a fusion prediction result of air quality; Based on the fusion prediction result and the detection point position, the position of a potential pollution source is marked, and a diffusion path simulation result is obtained based on a diffusion model, and based on the fusion prediction result, the position of the potential pollution source and the diffusion path simulation result, a real-time dynamic control strategy is generated to control the operating state of the fresh air system; wherein the real-time dynamic control strategy includes target fresh air volume, FFU adjustment range and FFU wind speed adjustment amount, which dilutes the pollutant concentration by adjusting the fresh air volume, and guides the airflow direction by adjusting the FFU adjustment range and the wind speed adjustment amount to accelerate the discharge of pollutants; Wherein, the marking of the potential pollution source position based on the fusion prediction result and the detection point position comprises: According to the fusion prediction result and the detection point position, the pollution source area is determined by spatial gradient analysis; A Gaussian mixture model is used to establish a source point feature distribution, and the detection point pollutant concentration and the detection point position are input to calculate the probability of each detection point position belonging to the source point cluster. When the probability is greater than a preset threshold, the detection point position is marked as the potential pollution source position; Wherein, the determination of the pollution source area according to the fusion prediction result and the detection point position through spatial gradient analysis comprises: The pollutant concentration of each detection point in the fusion prediction result is interpolated as a continuous distribution, and the gradient and Laplacian value are calculated according to the interpolation result to determine the spatial variation trend; Mark the grid with local peak value and outward radiation change trend as the candidate source; For each candidate source, backtrack multiple time points along the wind direction and negative gradient direction combined with wind speed and pressure difference to determine the pollution source area according to the change trend threshold, peak value threshold and consistency threshold. According to the diffusion model, the diffusion path simulation result is obtained, comprising:

2. The clean room fresh air system management method according to claim 1, wherein, According to the diffusion model, the predicted pollutant concentration of each detection point at different time points is calculated to construct a predicted pollutant concentration sequence; According to the predicted pollutant concentration sequence of each detection point, the diffusion path simulation result is obtained; Wherein, the diffusion model formula: Before the marking of the potential pollution source position based on the fusion prediction result and the detection point position, it further comprises: , representing a location at a time predicted pollutant concentration, is an initial pollutant concentration, is a diffusion coefficient, is a distance from a pollution source to a location .

3. The clean room fresh air system management method according to claim 1, wherein, According to the fusion prediction result, it is judged whether there is an index with concentration rising above a set threshold among the particle concentration, total acidic pollutant TA concentration, total basic pollutant TB concentration and VOC concentration; Correspondingly, when there is an index with concentration rising above a set threshold, the operation of marking the potential pollution source position based on the fusion prediction result and the detection point position and the subsequent operation is performed; otherwise, the fresh air system remains in the current state. ​ 4. The clean room fresh air system management method of claim 1, wherein, After the generation of the real-time dynamic control strategy to control the operation state of the fresh air system, the method further comprises: updating the acquired air quality data, energy consumption of the fresh air system and the purification equipment; inputting the real-time collected air quality data and energy consumption of the fresh air system and the purification equipment into the trained deep reinforcement learning network to obtain an optimized control strategy of the fresh air system; wherein the deep reinforcement learning network training process is as follows: designing a state space S of deep reinforcement learning, S comprising real-time air quality parameters of the clean room, control parameters of the real-time dynamic control strategy and system energy consumption data; setting an action space A, A comprising action modes a of fresh air volume increase and decrease, FFU air speed adjustment and FFU switch state switching; constructing a reward function: wherein, is the energy consumption of the fresh air system and the purification equipment; , is the weight coefficient of the reward function, indicating the weight of the energy consumption of the fresh air system and the purification equipment and the air quality parameters in the reward, is the air quality compliance rate; building a deep reinforcement learning framework based on the PPO algorithm, extracting S, A and R according to historical data in the clean room, training the deep reinforcement learning framework to obtain a trained deep reinforcement learning network; wherein the historical data in the clean room comprises air quality parameters, energy consumption of the fresh air system and the purification equipment, target fresh air volume, FFU adjustment range and FFU air speed adjustment amount.

5. An electronic device, comprising: a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the method according to any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that, a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the method according to any one of claims 1 to 4.

7. A computer program product, characterised in that, a computer program, wherein the computer program is executed by a processor to implement the method according to any one of claims 1 to 4.

Images