Method and system for controlling environmental parameters of an electronic clean room

Abstract

The application discloses a kind of electronic clean workshop environmental parameter control method and system, and the application relates to electronic clean workshop technical field.The method steps include: the workshop is divided into several clean workshop subareas and collects clean workshop subarea environment data, carries out fluctuation analysis and morphological analysis respectively to obtain subarea environment fluctuation value and subarea length-width ratio, to calculate air supply mode index to determine air supply mode;Real-time collection clean workshop subarea environment data input long short-term memory network model data, output particulate matter concentration prediction value sequence;Collect historical particulate matter concentration sequence, combine prediction value sequence to form comprehensive particulate matter concentration sequence, and obtain particulate matter concentration curve using least square method fitting;According to static pressure value, calculate the static pressure deviation of adjacent subarea, based on air supply mode, build workshop static pressure adjustment model, obtain first air supply and first exhaust air volume;Based on particulate matter concentration curve, calculate self-purification time, correct first air supply and first exhaust air volume by self-purification time.

Description

Technical Field

[0001] This invention relates to the field of electronic cleanroom technology, specifically to a method and system for controlling environmental parameters in electronic cleanrooms. Background Technology

[0002] As a critical environment for high-end electronic manufacturing, such as semiconductor chips and precision electronic devices, cleanrooms in the electronics industry require precise control of environmental parameters, including temperature, humidity, particulate matter concentration, static pressure difference, and micro-vibration, which directly impact product yield and performance. With electronic manufacturing processes becoming increasingly sophisticated and integrated, the requirements for the stability of environmental parameters are becoming increasingly stringent. For example, in semiconductor lithography, temperature fluctuations must be controlled within ±0.1℃, and particulate matter concentration must be maintained at extremely low levels. Any minute environmental deviation can lead to device failure. Therefore, developing efficient and precise environmental parameter control technologies has become one of the core requirements for high-quality development in the electronics manufacturing industry.

[0003] In existing technologies, the environmental parameters of electronic cleanrooms are generally adjusted by static thresholds of the cleanroom parameters. That is, the air supply and exhaust volumes are adjusted by comparing the real-time monitored instantaneous parameters of the cleanroom (such as temperature) with preset thresholds.

[0004] However, in traditional methods, electronic cleanrooms are adjusted based on static thresholds, adjusting the air supply and exhaust volumes only according to a single instantaneous parameter monitored in real time. This does not take into account the dynamic changes in the sub-area environment and ignores the changing trends of environmental parameters, resulting in delayed or excessive air supply. Consequently, the air supply mode does not match the actual needs of the cleanroom and cannot meet the stringent requirements of high-precision electronic manufacturing for a stable environment. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for controlling environmental parameters in electronic cleanrooms, thereby resolving the problems existing in the background technology.

[0006] To achieve the above objectives, the present invention provides a method and system for controlling environmental parameters in an electronic cleanroom, comprising the following steps:

[0007] Step S1: Divide the electronic cleanroom into several cleanroom housing areas, install sensors in the cleanroom housing areas, and collect data from the cleanroom housing areas through the sensors to obtain the housing area environmental data.

[0008] Step S2: By performing fluctuation analysis on the environmental data of the factory building area, the environmental fluctuation value of the sub-area is calculated; by performing morphological analysis on the cleanroom building area, the aspect ratio of the sub-area is calculated.

[0009] Step S3: By combining the environmental fluctuation value of the sub-area and the aspect ratio of the sub-area, the air supply mode index is calculated, and the air supply mode index is compared with the preset air supply index threshold to determine the air supply mode of the cleanroom building area.

[0010] Step S4: After determining the air supply mode, input the environmental data of the factory building area into the Long Short-Term Memory Network model for training to obtain the trained Long Short-Term Memory Network model; collect the environmental data of the factory building area in real time and input it into the trained Long Short-Term Memory Network model to output the particulate matter concentration prediction value sequence.

[0011] Step S5: Collect historical particulate matter concentration data of the cleanroom building area to obtain a historical particulate matter concentration sequence; combine the predicted particulate matter concentration sequence and the historical particulate matter concentration sequence to obtain a comprehensive particulate matter concentration sequence; perform curve fitting on the comprehensive particulate matter concentration sequence using the least squares method to obtain a particulate matter concentration curve;

[0012] Step S6: Calculate the static pressure deviation between adjacent cleanroom buildings based on the static pressure values ​​in the environmental data of the cleanroom buildings; construct a static pressure adjustment model based on the air supply mode, input the static pressure deviation into the static pressure adjustment model, and output the first air supply volume and the first exhaust volume of the cleanroom buildings; calculate the self-cleaning time of the particulate matter concentration based on the particulate matter concentration curve; based on the self-cleaning time and a preset self-cleaning time threshold, correct the first air supply volume and the first exhaust volume through the self-cleaning time to obtain the second air supply volume and the second exhaust volume, thereby achieving control of environmental parameters.

[0013] Preferably, the step of calculating the sub-area environmental fluctuation value by performing fluctuation analysis on the environmental data of the factory building area includes the following specific steps:

[0014] By performing fluctuation analysis on the environmental data of the factory building area, the fluctuation value of each feature of the environmental data of the factory building area was calculated:

[0015]

[0016] in, Let N be the fluctuation value of the j-th feature in the factory building area environmental data, N be the window size, and k be the index variable. This represents the k-th actual measurement value of the j-th feature within the sliding window. Let be the target value of the j-th feature, and t be the time t. The maximum allowable static deviation for the j-th feature. For time intervals, This represents the maximum allowable rate of change for the j-th feature. For static deviation weights, Weighted by rate of change + =1;

[0017] By summing the environmental features in the factory building area data, we obtain the sub-area environmental fluctuation value:

[0018]

[0019] in, Here, j represents the environmental fluctuation value of the sub-region, j is the feature index, and m is the number of features in the factory building area environmental data. Let be the weight of the j-th feature in the factory building area environmental data. Let be the fluctuation value of the j-th feature in the factory building area environmental data.

[0020] Preferably, the step of calculating the aspect ratio of the sub-area by performing morphological analysis on the cleanroom building area includes the following steps:

[0021] In determining the spatial morphology of a cleanroom building in an electronic cleanroom, the length of a geometric region is specifically defined by its length-to-width ratio. This ratio is used to quantify the elongation or narrowness of the region and provides a spatial basis for switching air supply modes.

[0022]

[0023] in, Let L be the aspect ratio of the sub-area, L be the length of the longest side of the cleanroom building area, and W be the length of the shortest side of the cleanroom building area.

[0024] Preferably, the step of calculating the air supply mode index by combining the sub-region environmental fluctuation value and the sub-region aspect ratio includes the following steps:

[0025] The aspect ratio of the sub-region is mapped using a non-linear function:

[0026]

[0027] in, The aspect ratio of the sub-region is mapped to the value of the nonlinear function. The aspect ratio of the sub-region;

[0028] By combining the environmental fluctuation value of the sub-region and the aspect ratio of the sub-region, the air supply mode index is calculated:

[0029]

[0030] Where Mode is the air supply mode index. This represents the environmental fluctuation value of the sub-region. The aspect ratio of the sub-region is mapped to the value of the nonlinear function. The weight of the environmental fluctuation value of the sub-region. The weights of the aspect ratio of the sub-region in the mapping value of the nonlinear function. + =1.

[0031] Preferably, after determining the air supply mode, the environmental data of the factory building area is input into a long short-term memory network model for training to obtain a trained long short-term memory network model, including the following steps:

[0032] Environmental data from the factory building area was input into a Long Short-Term Memory (LSTM) network model for training. Temperature, humidity, illuminance, noise, and micro-vibration from the environmental data were used as independent variables, and particulate matter concentration was used as the dependent variable. The input independent and dependent variables were normalized using a minimum-maximum method to obtain normalized environmental data. This LSTM network model consists of an input layer, two LSTM layers, an attention layer, a fully connected layer, and an output layer. The input layer receives the normalized environmental data from the factory building area. The first LSTM layer contains 64 units, and the second LSTM layer contains 32 units. The attention layer uses the Bahdanau attention mechanism. The fully connected layer has two layers: the first layer contains 32 neurons, and the second layer contains 1 neuron. The output layer uses a linear activation function to output a sequence of predicted particulate matter concentration values. During training, the Adam optimizer was used with an initial learning rate of 0.001 and mean squared error as the loss function. Through continuous optimization and tuning, the trained LSTM network model was finally obtained.

[0033] Preferably, the real-time collection of environmental data from the factory building area and input into a trained long short-term memory network model to output a sequence of predicted particulate matter concentrations includes the following steps:

[0034] Real-time environmental data of the factory building area is collected and input into a trained long short-term memory network model. The output is a sequence of predicted particulate matter concentration values ​​for the next time interval T: [ , , ,..., ].

[0035] Preferably, the step of curve fitting the comprehensive particulate matter concentration sequence using the least squares method to obtain the particulate matter concentration curve includes the following specific steps:

[0036] The particulate matter concentration sequence was curve-fitted using the least squares method to obtain the particulate matter concentration curve:

[0037]

[0038] in, Here is the particulate matter concentration curve, where A is the initial peak value of the particulate matter concentration curve, k is the attenuation constant, and B is the background concentration.

[0039] Preferably, the step of constructing a static pressure regulation model for the plant based on the air supply mode includes the following specific steps:

[0040] Calculate the static pressure deviation between adjacent cleanroom building areas. The static pressure of each cleanroom building area is determined by a quadratic equation relating the supply air volume and the exhaust air volume.

[0041]

[0042] in, Let be the static pressure of the i-th cleanroom building area. Let be the static pressure airflow characteristic coefficient of the i-th cleanroom building area. Let be the air supply volume for the i-th cleanroom building area. Let N be the exhaust volume of the i-th cleanroom building area, and N be the number of cleanroom buildings.

[0043] Pressure differential constraint conditions for the static pressure regulation model of the plant: The pressure differential between adjacent cleanroom units must meet the process requirements. ;

[0044] Physical limitations of supply and exhaust air volume:

[0045]

[0046] in, Let be the air supply volume for the i-th cleanroom building area. Let be the exhaust volume of the i-th cleanroom building area. and For minimum and maximum air supply volume, and These are the minimum and maximum exhaust volumes;

[0047] Wind turbine capacity limitations:

[0048]

[0049] in, Let be the air supply volume for the i-th cleanroom building area. Let be the exhaust volume of the i-th cleanroom building area. N represents the total capacity of the fans in the electronic cleanroom, and N represents the number of cleanroom units.

[0050] Objective function of the plant static pressure regulation model:

[0051]

[0052] in, Let be the objective function. This represents the penalty factor for differential pressure constraints, with a value ranging from 0.5 to 4. The target static pressure of the area is represented by N, which is the number of cleanroom buildings.

[0053] Preferably, the step of correcting the first supply air volume and the first exhaust air volume through the self-cleaning time to obtain the second supply air volume and the second exhaust air volume includes the following specific steps:

[0054] The first air supply volume is corrected by the self-cleaning time to obtain the second air supply volume:

[0055]

[0056] in, The second air supply volume for the i-th cleanroom building area. Let i be the initial air supply volume for the i-th cleanroom building area. This is the airflow gain coefficient, with a default value of 0.5. To preset the self-cleaning time threshold, This is the time scale factor, with a default value of 2.5. Self-cleaning time;

[0057] The second exhaust volume is obtained by adjusting the first exhaust volume based on the second supply air volume:

[0058]

[0059] in, Let i be the second exhaust volume of the i-th cleanroom building area. Let i be the first exhaust volume of the i-th cleanroom building area. The second air supply volume for the i-th cleanroom building area. This is the exhaust compensation coefficient. Let i be the initial air supply volume for the i-th cleanroom building area.

[0060] A control system for environmental parameters of an electronic cleanroom includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0061] This invention provides a method for controlling environmental parameters in an electronic cleanroom, involving machine learning and deep learning technologies, which has the following beneficial effects:

[0062] (1) The air supply mode index is calculated by combining the environmental fluctuation value of the sub-zone and the length-width ratio of the sub-zone. The influence of environmental stability and spatial shape on airflow distribution is comprehensively considered. The environmental fluctuation value reflects the stability of the real-time environmental parameters of the sub-zone, while the length-width ratio reflects the constraint of the narrow and long characteristics of the space on the uniformity of airflow. The air supply mode index obtained by weighted calculation can be used to target different sub-zones such as the core process area, which pays more attention to fluctuation, and the narrow auxiliary area, which pays more attention to the shape matching top supply and bottom exhaust, side supply and bottom exhaust or mixed air supply, which not only ensures the accuracy of environmental control, but also improves energy saving and applicability.

[0063] (2) Combining the predicted particulate matter concentration sequence and the historical particulate matter concentration sequence to obtain a comprehensive particulate matter concentration sequence is significant because it takes into account both historical patterns and future trends. The historical sequence provides the past characteristics of particulate matter concentration changes, while the predicted sequence supplements the future development trend, avoiding the lag of relying solely on real-time data.

[0064] (3) The first supply air volume and the first exhaust air volume are corrected by the self-cleaning time to obtain the second supply air volume and the second exhaust air volume, realizing precise control of dynamic adaptation. The first supply and exhaust air volume is determined based on the static pressure regulation model to ensure static pressure balance, while the self-cleaning time reflects the actual rate of decrease in particulate matter concentration. By comparing the self-cleaning time with the preset self-cleaning time threshold, the supply air volume and exhaust air volume are corrected to ensure that the particulate matter concentration in the electronic cleanroom always meets the production requirements. Attached Figure Description

[0065] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0066] Figure 1 This is a flowchart of the steps for controlling environmental parameters in an electronic cleanroom proposed in this invention.

[0067] Figure 2 This is a step hierarchy diagram of obtaining the air supply mode index in the method for controlling environmental parameters of an electronic cleanroom proposed in this invention;

[0068] Figure 3 This is a step hierarchy diagram of obtaining the particulate matter concentration curve in the method for controlling environmental parameters of an electronic cleanroom proposed in this invention. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Please see Figures 1-3 The present invention provides a technical solution: a method for controlling environmental parameters in an electronic cleanroom.

[0071] Step S1: Divide the electronic cleanroom into several cleanroom zones, install sensors in the cleanroom zones, and collect data from the cleanroom zones through the sensors to obtain environmental data of the cleanroom zones.

[0072] The electronic cleanroom is divided into several cleanroom zones based on their core functions and precision requirements. Specifically, these zones are divided into core process zones (such as areas responsible for high-precision processing like semiconductor chip lithography and thin-film deposition), auxiliary operation zones (such as areas responsible for wafer transfer and equipment maintenance), buffer transition zones (such as changing areas and material handling rooms responsible for environmental isolation and transition), and utility equipment zones (such as air conditioning rooms and power stations providing energy supply). This ensures that the division of each sub-zone is compatible with its actual production responsibilities and required environmental control precision. Different types of high-sensitivity sensors are installed at key production stations, air supply terminals, return air ducts, inside equipment housings, and along the boundaries of the enclosure structure within the cleanroom zones. These sensors include high-precision temperature sensors, high-precision humidity sensors, illuminance sensors, laser particle counters, differential pressure sensors, micro-vibration sensors, and electromagnetic field strength meters, enabling real-time monitoring of all parameters such as temperature, humidity, particulate matter concentration, illuminance, static pressure difference, and micro-vibration.

[0073] Intelligent data fusion algorithms are used to denoise, weight confidence levels, and remove anomalies from different parameters, forming a high-precision environmental status database to provide accurate data for subsequent control. Kalman filtering is employed to eliminate cross-interference between sensors, and data confidence weights are dynamically adjusted based on historical calibration data. An isolated forest algorithm is used to identify data exceeding thresholds, and its effectiveness is verified by combining spatiotemporal correlation. Finally, edge computing nodes are deployed for data denoising and normalization preprocessing, ultimately yielding the environmental data for the factory building area.

[0074] It should be noted that in electronic cleanrooms, temperature changes affect the intensity of air convection, thereby altering the diffusion and settling rates of particulate matter. For example, airflow created by temperature gradients may drive particulate matter to migrate between different areas. Humidity directly affects the agglomeration state of particulate matter. In high humidity environments, tiny particles are more likely to increase in size due to water vapor condensation, accelerating settling, while low humidity may make particles more likely to remain suspended. Humidity also affects the adsorption efficiency of filter media in air filtration systems. Illumination changes are often related to the start-up and shutdown of production equipment and personnel operations, which are often accompanied by the generation and diffusion of particulate matter. Noise is mostly caused by equipment operation (such as fans and precision instruments). The noise sources generated by equipment operation are usually accompanied by airflow disturbances, which alter the local airflow field and affect the uniformity of particulate matter distribution. Micro-vibrations come from equipment vibration, ground conduction, etc. Vibrations can resuspend particulate matter deposited on the surface, increasing the concentration of particulate matter in the air. At the same time, air fluctuations caused by vibrations can also interfere with the stable settling process of particulate matter.

[0075] Step S2: By performing fluctuation analysis on the environmental data of the factory building area, the environmental fluctuation value of the sub-area is calculated; by performing morphological analysis on the clean factory building area, the aspect ratio of the sub-area is calculated.

[0076] In the calculation of environmental fluctuation values ​​in sub-regions, the sliding window method is a very suitable choice for processing sampled data. It can dynamically retain "real-time data of the most recent period" to avoid lag caused by too much historical data. At the same time, it balances the calculation stability and real-time response speed by using a fixed window size, which is very suitable for scenarios that require real-time control (such as air supply mode switching and air volume static pressure coupling control).

[0077] Using a sliding window method, a fixed window size of N (containing the most recent N sampled data points) is set, and the sampling interval is... If sampling occurs every minute, then the window coverage time is N*. Each new data point collected is denoted as t represents the current time, and the window automatically removes the oldest data. ,reserve[ , ,..., There are a total of N latest data points.

[0078] For each feature sequence in the factory building environmental data, the fluctuation value of each environmental parameter (or feature) in the factory building environmental data is calculated by performing fluctuation analysis on the data:

[0079]

[0080] in, Let N be the fluctuation value of the j-th feature in the factory building area environmental data, N be the window size, and k be the index variable. This represents the k-th actual measurement value of the j-th feature within the sliding window. Let be the target value of the j-th feature, and t be the time t. The maximum allowable static deviation for the j-th feature. For time intervals, This represents the maximum allowable rate of change for the j-th feature. For static deviation weights, Weighted by rate of change + =1.

[0081] It should be noted that the fluctuation values ​​of each environmental parameter (or feature) in the sub-region are categorized into two terms. The first term is the static deviation term, which quantifies the overall deviation of the parameter from the target value by calculating the average absolute deviation of each measured value within the sliding window from the target value and then dividing it by the maximum allowable static deviation of the parameter. The second term is the rate of change term, which captures the most drastic fluctuations of the parameter over a short period by extracting the maximum rate of change of adjacent measured values ​​within the sliding window and then dividing it by the maximum allowable rate of change of the parameter. These two terms are weighted by a coefficient. and Weighted summation is performed to comprehensively reflect the overall stability of the parameters and the risk of instantaneous fluctuations.

[0082] It should be noted that the maximum static deviation and maximum rate of change for each characteristic in the factory building environmental data are determined based on the physical characteristics and control precision requirements of each characteristic. For example, in the core process area of ​​an electronic cleanroom, the maximum allowable static deviation for temperature is 23.0 ± 0.1℃, and the maximum allowable rate of change for temperature is ≤ 0.05℃ / min; the maximum allowable static deviation for humidity is 45 ± 2%RH, and the maximum allowable rate of change for humidity is ≤ 0.1%RH / min. The maximum allowable static deviation for particulate matter concentration is ≤ 10%, and the maximum allowable rate of change for particulate matter concentration is ≤ 2 particles / m³. 3 •min (particle concentration increase not exceeding 2 particles per minute); Maximum allowable static deviation of wind speed: 0.45±0.02m / s, Maximum allowable rate of change of wind speed: ≤0.01m / s·min (wind speed fluctuation not exceeding 0.01m / s per minute). Maximum allowable static deviation of static pressure difference: ≥8Pa±0.5Pa, Maximum allowable rate of change of static pressure difference: ≤0.3Pa / min (pressure difference fluctuation not exceeding 0.3Pa per minute).

[0083] It should be noted that the static deviation weight and rate of change weight If the cleanroom building area is the core process area, then environmental stability is the priority, and in this case, it is necessary to increase the size of the cleanroom. , 0.7 is acceptable. The value is 0.3; if the cleanroom area is an auxiliary operation area, the risk of mutation is high and the value needs to be increased. , 0.7 is acceptable. The value is 0.3; if the cleanroom building area is a buffer zone, environmental stability is a priority. 0.6 is acceptable. The weight is 0.4; if the cleanroom building area is a public equipment area, then the weight is balanced. Take 0.5, Take 0.5.

[0084] By summing the environmental features in the factory building area data, we obtain the sub-area environmental fluctuation value:

[0085]

[0086] in, Here, j represents the environmental fluctuation value of the sub-region, j is the feature index, and m is the number of features in the factory building area environmental data. Let be the weight of the j-th feature in the factory building area environmental data. Let be the fluctuation value of the j-th feature in the factory building area environmental data.

[0087] It should be noted that the weight of the j-th feature in the factory building area environmental data... The weighting depends on the type of cleanroom building area. For example, if the cleanroom building area is the core process area, the fluctuation value of particulate matter concentration is particularly important, followed by temperature and humidity. In this case, the weight of particulate matter concentration should be increased to 0.4, the weight of temperature to 0.2, the weight of humidity to 0.15, and the total weight of other parameters to 0.25.

[0088] In determining the spatial morphology of a cleanroom building in an electronic cleanroom, the length of a geometric region is specifically defined by its length-to-width ratio. This ratio is used to quantify the elongation or narrowness of the region and provides a spatial basis for switching air supply modes.

[0089]

[0090] in, Let L be the aspect ratio of the sub-area, L be the length of the longest side of the cleanroom building area, and W be the length of the shortest side of the cleanroom building area.

[0091] It should be noted that in the description of the planar form of the sub-areas in the factory building cleanroom, the length L of the longest side and the length W of the shortest side are distinguished according to the geometric dimensions of the regional plane by numerical values. The core is to judge whether the space is open or narrow through the aspect ratio AR. The length L of the longest side refers to the length of the side with the longest distance between two opposite sides in the plane of the sub-area of the cleanroom. For example, if the area is a rectangle of 5m longitudinally × 3m transversely, the longest side L = 5m (because 5m > 3m); if the area is an approximate square of 4m × 4.2m, the longest side L = 4.2m (take the side with a larger value); even if the area is not a standard rectangle (such as a slightly irregular operation area), it can be simplified to an equivalent rectangle (the largest inscribed rectangle), and the length of the two opposite sides with the farthest distance in its plane is taken as L. The length W of the shortest side is the length of the side with the shortest distance between two opposite sides in the plane of the sub-area of the cleanroom (that is, the side with a numerical value less than or equal to the longest side). For example, in a 5m × 3m area, the shortest side W = 3m; in a 4m × 4.2m area, the shortest side W = 4m; after the irregular area is simplified to an equivalent rectangle, the length of the opposite side with a smaller value is taken as W.

[0092] Step S3: By combining the sub-area environmental fluctuation value and the aspect ratio of the sub-area, calculate the air supply mode index, compare the air supply mode index with a preset air supply index threshold, and determine the air supply mode of the sub-area of the cleanroom.

[0093] Perform a non-linear function mapping on the aspect ratio of the sub-area:

[0094]

[0095] Among them, The mapped value of the aspect ratio of the sub-area in the non-linear function, is the aspect ratio of the sub-area.

[0096] It should be noted that for the non-linear function of the aspect ratio of the sub-area, AR = 1 represents a square area. The air flow distribution in the square area is the most uniform, and there are no significant air flow dead corners. f(AR) = 0 indicates that the shape has no additional influence on the air flow uniformity; 1 < AR ≤ 2 (medium and narrow area), as the aspect ratio increases (1 < AR ≤ 2), the area gradually becomes narrower and longer, and there may be slight air flow dead corners (such as in the middle of the long side), and f(AR) = 0.5(AR - 1) increases linearly, reflecting that the influence of the shape on the air flow uniformity increases with the increase of the narrow and long degree; AR > 2 (extremely narrow and long area), when the aspect ratio exceeds 2, the area is extremely narrow and long, and it is difficult for the air flow to cover the entire cross-section, and there must be significant air flow dead corners. f(AR) = 0.5 (saturation value), indicating that the influence of the shape on the air flow uniformity has reached the upper limit, and forced compensation through the air supply mode is required (such as increasing local pressurization).

[0097] By combining the sub-area environmental fluctuation value and the aspect ratio of the sub-area, calculate the air supply mode index:

[0098]

[0099] Among them, Mode is the air supply mode index, is the environmental fluctuation value of the sub-region, is the mapping value of the aspect ratio of the sub-region in the non-linear function, is the weight of the environmental fluctuation value of the sub-region, is the weight of the mapping value of the aspect ratio of the sub-region in the non-linear function, + = 1.

[0100] It should be noted that the weight of the environmental fluctuation value of the sub-region and the weight of the mapping value of the aspect ratio of the sub-region in the non-linear function , for the core process area (such as the semiconductor chip lithography area), environmental fluctuations have a greater impact on production, can take values from 0.6 to 0.8, correspondingly take 0.4 to 0.2; for the auxiliary area with a significant degree of elongation (such as the long strip-shaped transmission channel), the spatial form has a more critical impact on the air flow distribution, can take values from 0.6 to 0.8, correspondingly take 0.4 to 0.2.

[0101] It should be noted that for the air supply mode index Mode, when Mode < 0.3, it indicates that the environmental fluctuation value of the sub-region fluctuates little and the shape of the sub-region in the clean factory is regular (such as square), and the top supply and bottom exhaust (unidirectional flow) method is adopted, without complex air supply, energy-saving and meeting the clean requirements; when 0.3 ≤ Mode < 0.7, it indicates that the environmental fluctuation value of the sub-region fluctuates moderately or the shape of the sub-region in the clean factory is moderately long and narrow (1 < R ≤ 2), and the side supply and bottom exhaust (non-unidirectional flow) is adopted, and the air flow disturbance needs to be enhanced to suppress the fluctuation, and the side supply and bottom exhaust can cover more areas; when Mode ≥ 0.7, it indicates that the environmental fluctuation value of the sub-region fluctuates greatly and the shape of the sub-region in the clean factory is extremely long and narrow (R > 2), and the mixed air supply and local pressurization are adopted to compensate for the air flow in the extremely long and narrow area.

[0102] Step S4: After determining the air supply mode, input the environmental data of the factory sub-region into the long short-term memory network model for training to obtain a trained long short-term memory network model; collect the environmental data of the factory sub-region in real time and input it into the trained long short-term memory network model, and output a sequence of predicted particulate matter concentration values.

[0103] When inputting factory building environmental data into a Long Short-Term Memory (LSTM) network model for training, it is first necessary to define the input and output variables: temperature, humidity, illuminance, noise, and micro-vibration from the factory building environmental data are used as independent variables, and particulate matter concentration is used as the dependent variable, ensuring that the model can learn the correlation between multi-dimensional environmental parameters and particulate matter concentration. To eliminate the influence of differences in the dimensions of different parameters on model training, the input independent and dependent variables need to be normalized to their maximum and minimum values, mapping all data to the [0,1] interval. This Long Short-Term Memory (LSTM) network model consists of an input layer, two LSTM layers, an attention layer, a fully connected layer, and an output layer. The input layer receives normalized environmental data from the factory building area. The first LSTM layer contains 64 units and returns the hidden state at each time step, providing the basis for the subsequent attention mechanism. The second LSTM layer contains 32 units and only outputs the hidden state at the last time step. The attention layer uses the Bahdanau attention mechanism, which calculates the weights of the hidden states at each time step and sums them by weight, focusing on the time steps with the greatest impact on particulate matter concentration (such as recent periods of high pollution) and key features (such as sudden changes in the first air supply volume) to obtain a context vector. The fully connected layer consists of two layers, with the first layer containing 32 neurons and the second layer containing 1 neuron. The output layer uses a linear activation function, and since particulate matter concentration is a continuous value, it ultimately outputs a sequence of predicted particulate matter concentration values. During training, the Adam optimizer was used with an initial learning rate of 0.001. An exponential decay strategy (decay rate of 0.95 per 10 epochs) was used to adaptively adjust the parameters and update the step size. The mean squared error (MSE) was used as the loss function (i.e., the sum of the squares of the differences between the predicted particulate matter concentration and the actual value output by the model). Through continuous optimization and tuning, a well-trained long short-term memory network model was finally obtained, which can accurately predict particulate matter concentration based on real-time input environmental data.

[0104] The particulate matter concentration from the factory building area environmental data was used as the dependent variable, while temperature, humidity, illuminance, noise, and micro-vibration were used as independent variables and input into the Long Short-Term Memory (LSTM) network model. The input independent variables (temperature, humidity, illuminance, noise, and micro-vibration) and dependent variable (particulate matter concentration) were normalized to their minimum and maximum values, mapping the data to the [0,1] interval to eliminate the influence of differences in the dimensions of different parameters on model training.

[0105] The Long Short-Term Memory (LSTM) network model comprises an input layer, two LSTM layers, an attention layer, a fully connected layer, and an output layer. It employs the Adam optimizer with an initial learning rate of 0.001 and uses an exponential decay strategy (decay rate 0.95 per 10 epochs) to adaptively adjust the parameter update step size.

[0106] Input layer: m features from the factory building area environmental data, each feature containing N data points within a time window. , ,..., ].

[0107] LSTM layer: First LSTM layer: 64 units, returns the complete sequence: outputs the hidden state at each time step, used for subsequent attention mechanisms.

[0108] The second LSTM layer has 32 cells and only returns the last time step.

[0109] Attention layer: Using the Bahdanau attention mechanism, the weights of the hidden states at each time step are calculated and summed to obtain the context vector.

[0110] Fully connected layer number: 1~2 layers, the first layer has 32 neurons, the second layer has 1 neuron.

[0111] Activation function: Linear activation is used for the output layer (because the particulate matter concentration is a continuous value).

[0112] Output layer: Outputs the predicted particulate matter concentration for the corresponding future time point.

[0113] By using mini-batch gradient descent, the factory building environment data is divided into B batches. The loss function of the Long Short-Term Memory (LSTM) network model is then:

[0114]

[0115] Where LOSS is the loss function, B is the batch size, T is the predicted sequence length, and b is the sample index within the batch. b B, k is the time point index within the predicted sequence, 1 k T, The sampling interval is... Indicates that the b-th sample is in The predicted value at time t Indicates that the b-th sample is in The actual value at time t.

[0116] Real-time environmental data of the factory building area is collected and input into a trained long short-term memory network model. The output is a sequence of predicted particulate matter concentration values ​​for the next time interval T: [ , , ,..., ].

[0117] It should be noted that long short-term memory network models can also be used to predict the fluctuation trends of parameters such as temperature, humidity, illuminance, noise, and micro-vibration, allowing for the early activation or adjustment of compensation equipment. For example, auxiliary dehumidification modules can be used to quickly reduce humidity fluctuations, adjustable lighting systems can maintain uniform illuminance, and vibration-damping support platforms can absorb the micro-vibrations generated during equipment operation.

[0118] Step S5: Collect historical particulate matter concentration data of the cleanroom building area to obtain a historical particulate matter concentration sequence; combine the predicted particulate matter concentration sequence and the historical particulate matter concentration sequence to obtain a comprehensive particulate matter concentration sequence; perform curve fitting on the comprehensive particulate matter concentration sequence using the least squares method to obtain a particulate matter concentration curve.

[0119] Historical particulate matter concentration data of the cleanroom building area were collected to obtain a historical particulate matter concentration sequence, wherein the historical particulate matter concentration sequence is a historical particulate matter concentration sequence with a time window of N: [ , ,..., ].

[0120] Historical particulate matter concentration sequences and predicted particulate matter concentration sequences are spliced ​​together in chronological order to form a comprehensive particulate matter concentration sequence.

[0121] The particulate matter concentration sequence was curve-fitted using the least squares method to obtain the particulate matter concentration curve:

[0122]

[0123] in, Here is the particulate matter concentration curve, where A is the initial peak value of the particulate matter concentration curve, k is the attenuation constant, and B is the background concentration.

[0124] It should be noted that the core significance of the particulate matter concentration curve obtained by fitting using the least squares method lies in the dynamic quantification of the purification process. This curve integrates historical monitoring data with future trends predicted by the Long Short-Term Memory Network, accurately depicting the change in particulate matter concentration over time in an exponential decay manner: the initial high concentration (characterized by A) decreases exponentially under the action of the purification system, eventually approaching the background concentration B. This fitting not only overcomes the limitations of traditional static threshold control but also reveals the actual effectiveness of the purification rate (represented by the decay constant k).

[0125] It should be noted that the initial peak value A represents the initial increase in particulate matter concentration relative to the background concentration at the moment the dust-generating event is triggered (such as sudden disturbances like equipment start-up and shutdown, material handling, or process operations). It reflects the instantaneous release intensity of the dust source (such as the dust burst force when equipment starts up or the initial concentration peak of material splashing). The background concentration B represents the inherent particulate matter concentration in the factory area after sufficient diffusion / settling without significant instantaneous dust generation interference. It is the baseline value for a clean environment (such as the steady-state concentration during continuous operation of high-efficiency air supply or the natural equilibrium concentration when the area is undisturbed).

[0126] Step S6: Calculate the static pressure deviation between adjacent cleanroom buildings based on the static pressure values ​​in the environmental data of the cleanroom buildings; construct a static pressure adjustment model based on the air supply mode, input the static pressure deviation into the static pressure adjustment model, and output the first air supply volume and the first exhaust volume of the cleanroom buildings; calculate the self-cleaning time of the particulate matter concentration based on the particulate matter concentration curve. If the self-cleaning time is less than a preset self-cleaning time threshold, the first air supply volume and the first exhaust volume are corrected based on the self-cleaning time to obtain the second air supply volume and the second exhaust volume, thereby achieving control of environmental parameters.

[0127] Calculate the static pressure deviation between adjacent cleanroom building areas. The static pressure of each cleanroom building area is determined by a quadratic equation relating the supply air volume and the exhaust air volume.

[0128]

[0129] in, Let be the static pressure of the i-th cleanroom building area. Let be the static pressure airflow characteristic coefficient of the i-th cleanroom building area. Let be the air supply volume for the i-th cleanroom building area. Let N be the exhaust volume of the i-th cleanroom building area, and N be the number of cleanroom buildings.

[0130] It should be noted that the static pressure air volume characteristic coefficient reflects the air duct resistance characteristics of the cleanroom building, such as the number of HEPA filters and the length of the duct. The greater the resistance, the more necessary it is to be calibrated experimentally. The range is 0.001-0.01.

[0131] Construct a static pressure regulation model for the plant:

[0132] Pressure difference constraints for the static pressure regulation model of the plant: The pressure difference between adjacent cleanroom units must meet process requirements. .

[0133] It should be noted that the pressure difference between adjacent areas, for example: This indicates that the static pressure of the i-th cleanroom building area is 7 Pa higher than that of the j-th cleanroom building area.

[0134] Physical limitations of supply and exhaust air volume:

[0135]

[0136] in, Let be the air supply volume for the i-th cleanroom building area. Let be the exhaust volume of the i-th cleanroom building area. and For minimum and maximum air supply volume, and These are the minimum and maximum exhaust volumes.

[0137] It should be noted that different air supply modes correspond to different physical limitations on supply and exhaust air volumes. For example, the top supply and bottom exhaust mode... For 5000 / h, For 8000 / h, For 5000 / h, For 8000 / h; Side-discharge bottom-outlet mode For 3000 / h, For 6000 / h, 2700 / h, 5400 / h, mixed air supply and localized pressurization mode For 5000 / h, For 10000 / h, For 6000 / h, 12000 / h.

[0138] Wind turbine capacity limitations:

[0139]

[0140] in, Let be the air supply volume for the i-th cleanroom building area. Let be the exhaust volume of the i-th cleanroom building area. N represents the total capacity of the fans in the electronic cleanroom, and N represents the number of cleanroom buildings.

[0141] Objective function of the plant static pressure regulation model:

[0142]

[0143] in, Let be the objective function. This represents the penalty factor for differential pressure constraints, with a value ranging from 0.5 to 4. The target static pressure of the area is represented by N, which is the number of cleanroom buildings.

[0144] The objective function of the cleanroom static pressure regulation model is minimized using the Lagrange multiplier method, ultimately yielding the initial air supply volume for each cleanroom building area. and the first row air volume .

[0145] It should be noted that replaceable multi-stage purification units, including particle filtration, high-efficiency molecular adsorption, and chemical reaction catalysis modules, are installed above the air supply ducts and working area to achieve targeted removal of acidic, alkaline, and organic molecular pollutants. Simultaneously, electrostatic neutralization devices and local EMI shielding units are installed in the working area and on the equipment casing to dissipate static electricity accumulation and electromagnetic interference in real time. The control system dynamically adjusts the airflow of the purification units and the ion release frequency based on pollutant concentration and electrostatic field strength, ensuring that micro-molecular pollutants and electrostatic / EMI indicators remain below threshold levels.

[0146] Based on the particulate matter concentration curve, from the current moment Initially, the concentration decreased to the safe threshold for particulate matter concentration. Time required:

[0147]

[0148] in, Self-cleaning time, for The particulate matter concentration at time t, where B is the background concentration and k is the attenuation constant. This is the safe threshold for particulate matter concentration.

[0149] It should be noted that, < When , it means that the numerator is greater than the denominator (both are positive numbers). >1, If it is a positive number, then A negative number indicates that the particulate matter concentration has already reached the standard. In this case, there is no need to enhance purification; instead, consider reducing the air volume (energy saving).

[0150] according to and Size, To preset the self-cleaning time threshold, >0, and the first air supply volume is corrected by the self-cleaning time to obtain the second air supply volume:

[0151]

[0152] in, The second air supply volume for the i-th cleanroom building area. Let i be the initial air supply volume for the i-th cleanroom building area. This is the airflow gain coefficient, with a default value of 0.5. To preset the self-cleaning time threshold, This is the time scale factor, with a default value of 2.5. This is the self-cleaning time.

[0153] It should be noted that the preset self-cleaning time threshold The core process area is 5 minutes, while the auxiliary operation area, buffer transition area, and common equipment area can be 10 minutes.

[0154] It should be noted that when When the self-cleaning time exceeds a certain threshold, it indicates that the purification time has expired, requiring an increase in airflow to improve purification efficiency (accelerate particulate matter discharge) and shorten the self-cleaning time to within the target range; when When the airflow is too fast, it indicates that the airflow needs to be reduced to lower energy consumption, while also avoiding airflow turbulence caused by excessive airflow.

[0155] The second exhaust volume is obtained by adjusting the first exhaust volume based on the second supply air volume:

[0156]

[0157] in, Let i be the second exhaust volume of the i-th cleanroom building area. Let i be the first exhaust volume of the i-th cleanroom building area. The second air supply volume for the i-th cleanroom building area. This is the exhaust compensation coefficient. Let i be the initial air supply volume for the i-th cleanroom building area.

[0158] It should be noted that the exhaust compensation coefficient bz has several implications. For example, in the top-supply, bottom-exhaust mode, the airflow is highly unidirectional and the coordination between supply and exhaust is crucial, so bz can be approximately 1. In the side-supply, bottom-exhaust mode, the airflow disturbance is greater, so the value of bz can be adjusted appropriately. When it is necessary to enhance the exhaust response to changes in supply air (such as accelerating the discharge of particulate matter generated by local disturbances), bz can be greater than 1, with a value of 1.05-1.2. When it is necessary to weaken the exhaust adjustment to avoid exacerbating airflow turbulence, bz can be less than 1, with a value of 0.8-0.9. In the mixed supply and local pressurization mode, if the local pressurization area (such as near the dust generation point) needs to enhance the exhaust response to changes in supply air to quickly discharge high-concentration particulate matter, the value of bz can be 1.1-1.3 (greater than 1). If the overall area needs to maintain stable static pressure to avoid overall airflow imbalance due to local pressurization, the value of bz can be 0.9-1.1 (close to 1).

[0159] Furthermore, based on the above method embodiments, the present invention also provides a control system, including a memory, a processor, and a computer program stored in the memory, which is adapted to be loaded and executed by the processor to implement the above-described method for controlling environmental parameters of an electronic cleanroom.

[0160] This paper proposes a method for controlling environmental parameters in electronic cleanrooms. This method divides the cleanroom into sub-areas such as core process areas and auxiliary operation areas, and deploys sensors to collect environmental data. The environmental fluctuation values ​​of the sub-areas are obtained through fluctuation analysis, and the aspect ratio of the sub-areas is obtained through morphological analysis. The air supply mode is determined by combining the two. Then, a long short-term memory network model is used to predict particulate matter concentration. The historical and predicted sequence fitting curves are fused, and finally, the air supply and exhaust volume is corrected by the self-cleaning time to achieve precise control of environmental parameters.

[0161] The significance of calculating the air supply mode index by combining the environmental fluctuation value and the aspect ratio of the sub-zone lies in comprehensively considering the impact of environmental stability and spatial morphology on airflow distribution. The environmental fluctuation value reflects the stability of the real-time environmental parameters of the sub-zone, while the aspect ratio reflects the constraint of the narrow and elongated characteristics of the space on the uniformity of airflow. The air supply mode index obtained by weighted calculation can be used to match top supply and bottom exhaust, side supply and bottom exhaust, or mixed air supply modes for different sub-zones (such as the core process area which focuses more on fluctuation, and the narrow auxiliary area which focuses more on morphology), ensuring both the accuracy of environmental control and improving energy efficiency and applicability.

[0162] Combining predicted particulate matter concentration sequences with historical particulate matter concentration sequences yields a comprehensive particulate matter concentration sequence, which is significant because it takes into account both historical patterns and future trends. The historical sequence provides past characteristics of particulate matter concentration changes, while the predicted sequence supplements future development trends, avoiding the lag of relying solely on real-time data.

[0163] The second supply and exhaust air volumes are derived by adjusting the first supply and exhaust air volumes using the self-cleaning time. This process enables precise, dynamically adapted control. The first supply and exhaust air volumes are determined based on a static pressure regulation model, ensuring static pressure balance. The self-cleaning time reflects the actual rate of decrease in particulate matter concentration. By comparing the self-cleaning time with a preset self-cleaning time threshold, the supply and exhaust air volumes are adjusted to ensure that the electronic cleanroom (especially high-precision process areas) meets production requirements, thus improving the flexibility and reliability of environmental control.

[0164] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the phrase "comprising an element defined as..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0165] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for controlling environmental parameters in an electronic cleanroom, characterized in that: Includes the following steps: Step S1: Divide the electronic cleanroom into several cleanroom housing areas, install sensors in the cleanroom housing areas, and collect data from the cleanroom housing areas through the sensors to obtain the housing area environmental data. Step S2: By performing fluctuation analysis on the environmental data of the factory building area, the environmental fluctuation value of the sub-area is calculated; by performing morphological analysis on the cleanroom building area, the aspect ratio of the sub-area is calculated. Step S3: By combining the environmental fluctuation value of the sub-area and the aspect ratio of the sub-area, the air supply mode index is calculated, and the air supply mode index is compared with the preset air supply index threshold to determine the air supply mode of the cleanroom building area. Step S4: After determining the air supply mode, input the environmental data of the factory building area into the Long Short-Term Memory Network model for training to obtain the trained Long Short-Term Memory Network model; collect the environmental data of the factory building area in real time and input it into the trained Long Short-Term Memory Network model to output the particulate matter concentration prediction value sequence. Step S5: Collect historical particulate matter concentration data of the cleanroom building area to obtain a historical particulate matter concentration sequence; combine the predicted particulate matter concentration sequence and the historical particulate matter concentration sequence to obtain a comprehensive particulate matter concentration sequence; perform curve fitting on the comprehensive particulate matter concentration sequence using the least squares method to obtain a particulate matter concentration curve; Step S6: Calculate the static pressure deviation between adjacent cleanroom buildings based on the static pressure values ​​in the environmental data of the cleanroom buildings; construct a static pressure adjustment model based on the air supply mode, input the static pressure deviation into the static pressure adjustment model, and output the first supply air volume and the first exhaust air volume of the cleanroom buildings; calculate the self-cleaning time of the particulate matter concentration based on the particulate matter concentration curve; based on the self-cleaning time and a preset self-cleaning time threshold, correct the first supply air volume and the first exhaust air volume using the self-cleaning time to obtain the second supply air volume and the second exhaust air volume, thereby achieving control of environmental parameters; specifically: ； in, The second air supply volume for the i-th cleanroom building area. Let i be the initial air supply volume for the i-th cleanroom building area. This is the airflow gain coefficient, with a default value of 0.

5. To preset the self-cleaning time threshold, This is the time scale factor, with a default value of 2.

5. Self-cleaning time; The second exhaust volume is obtained by adjusting the first exhaust volume based on the second supply air volume: ； in, Let i be the second exhaust volume of the i-th cleanroom building area. Let i be the first exhaust volume of the i-th cleanroom building area. The second air supply volume for the i-th cleanroom building area. This is the exhaust compensation coefficient. Let i be the initial air supply volume for the i-th cleanroom building area.

2. The method for controlling environmental parameters in an electronic cleanroom according to claim 1, characterized in that: The step of calculating the sub-area environmental fluctuation value by performing fluctuation analysis on the environmental data of the factory building area includes the following specific steps: By performing fluctuation analysis on the environmental data of the factory building area, the fluctuation value of each feature of the environmental data of the factory building area was calculated: ； in, Let N be the fluctuation value of the j-th feature in the factory building area environmental data, N be the window size, and k be the index variable. This represents the k-th actual measurement value of the j-th feature within the sliding window. Let be the target value of the j-th feature, and t be the time t. The maximum allowable static deviation for the j-th feature. For time intervals, This represents the maximum allowable rate of change for the j-th feature. For static deviation weights, Weighted by rate of change + =1; By summing the environmental features in the factory building area data, we obtain the sub-area environmental fluctuation value: ； in, Here, j represents the environmental fluctuation value of the sub-region, j is the feature index, and m is the number of features in the factory building area environmental data. Let be the weight of the j-th feature in the factory building area environmental data. Let be the fluctuation value of the j-th feature in the factory building area environmental data.

3. The method for controlling environmental parameters in an electronic cleanroom according to claim 2, characterized in that: The step of calculating the aspect ratio of sub-areas by performing morphological analysis on the cleanroom building area includes the following steps: In determining the spatial morphology of a cleanroom building in an electronic cleanroom, the length of a geometric region is specifically defined by its length-to-width ratio. This ratio is used to quantify the elongation or narrowness of the region and provides a spatial basis for switching air supply modes. ； in, Let L be the aspect ratio of the sub-area, L be the length of the longest side of the cleanroom building area, and W be the length of the shortest side of the cleanroom building area.

4. The method for controlling environmental parameters in an electronic cleanroom according to claim 3, characterized in that: The process of calculating the air supply mode index by combining the environmental fluctuation value of the sub-region and the aspect ratio of the sub-region includes the following steps: The aspect ratio of the sub-region is mapped using a non-linear function: ； in, The aspect ratio of the sub-region is mapped to the value of the nonlinear function. The aspect ratio of the sub-region; By combining the environmental fluctuation value of the sub-region and the aspect ratio of the sub-region, the air supply mode index is calculated: ； Where Mode is the air supply mode index. This represents the environmental fluctuation value of the sub-region. The aspect ratio of the sub-region is mapped to the value of the nonlinear function. The weights of the sub-region environmental fluctuation values. The weights of the aspect ratio of the sub-region in the mapping value of the nonlinear function. + =1.

5. The method for controlling environmental parameters in an electronic cleanroom according to claim 4, characterized in that: After determining the air supply mode, the environmental data of the factory building area is input into the Long Short-Term Memory (LSTM) network model for training to obtain the trained LTM network model, including the following steps: Environmental data from the factory building area was input into a Long Short-Term Memory (LSTM) network model for training. Temperature, humidity, illuminance, noise, and micro-vibration from the environmental data were used as independent variables, and particulate matter concentration was used as the dependent variable. The input independent and dependent variables were normalized using a minimum-maximum method to obtain normalized environmental data. This LSTM network model consists of an input layer, two LSTM layers, an attention layer, a fully connected layer, and an output layer. The input layer receives the normalized environmental data from the factory building area. The first LSTM layer contains 64 units, and the second LSTM layer contains 32 units. The attention layer uses the Bahdanau attention mechanism. The fully connected layer has two layers: the first layer contains 32 neurons, and the second layer contains 1 neuron. The output layer uses a linear activation function to output a sequence of predicted particulate matter concentration values. During training, the Adam optimizer was used with an initial learning rate of 0.001 and mean squared error as the loss function. Through continuous optimization and tuning, the trained LSTM network model was finally obtained.

6. The method for controlling environmental parameters in an electronic cleanroom according to claim 5, characterized in that: The process of collecting real-time environmental data from the factory building area and inputting it into a trained long short-term memory network model to output a sequence of predicted particulate matter concentrations includes the following steps: Real-time environmental data of the factory building area is collected and input into a trained long short-term memory network model. The output is a sequence of predicted particulate matter concentration values ​​for the next time interval T: [ , , ,..., ].

7. The method for controlling environmental parameters in an electronic cleanroom according to claim 6, characterized in that: The step of curve fitting the comprehensive particulate matter concentration sequence using the least squares method to obtain the particulate matter concentration curve includes the following specific steps: The particulate matter concentration sequence was curve-fitted using the least squares method to obtain the particulate matter concentration curve: ； in, Here is the particulate matter concentration curve, where A is the initial peak value of the particulate matter concentration curve, k is the attenuation constant, and B is the background concentration.

8. The method for controlling environmental parameters in an electronic cleanroom according to claim 7, characterized in that: The process of constructing a static pressure regulation model for the plant based on the air supply mode includes the following specific steps: Calculate the static pressure deviation between adjacent cleanroom building areas. The static pressure of each cleanroom building area is determined by a quadratic equation relating the supply air volume and the exhaust air volume. ； in, Let be the static pressure of the i-th cleanroom building area. Let be the static pressure airflow characteristic coefficient of the i-th cleanroom building area. Let be the air supply volume for the i-th cleanroom building area. Let N be the exhaust volume of the i-th cleanroom building area, and N be the number of cleanroom buildings. Pressure difference constraints for the static pressure regulation model of the plant: The pressure difference between adjacent cleanroom units must meet process requirements. ; Physical limitations of supply and exhaust air volume: ； in, Let be the air supply volume for the i-th cleanroom building area. Let be the exhaust volume of the i-th cleanroom building area. and For minimum and maximum air supply volume, and These are the minimum and maximum exhaust volumes; Wind turbine capacity limitations: ； in, Let be the air supply volume for the i-th cleanroom building area. Let be the exhaust volume of the i-th cleanroom building area. N represents the total capacity of the fans in the electronic cleanroom, and N represents the number of cleanroom units. Objective function of the static pressure regulation model for a plant: ； in, Let be the objective function. This represents the penalty factor for differential pressure constraints, with a value ranging from 0.5 to 4. The target static pressure of the area is represented by N, which is the number of cleanroom buildings.

9. A control system for environmental parameters in an electronic cleanroom, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-8.

Images