A dynamic control method and system for exhaust air volume of a fume hood

Abstract

The present application relates to the field of exhaust air volume control, and particularly relates to a dynamic control method and system for exhaust air volume of a fume hood, which comprises the following steps: dividing a to-be-detected period into several sub-periods to obtain a target point in a target period, and calculating the disturbance degree of the target point; training a prediction model based on a feature sequence in history, constructing a detection window, inputting the feature sequence of the detection window into the prediction model to obtain a prediction window, and calculating the required air volume of the target point at a to-be-adjusted time; traversing to obtain the required air volume of each monitoring point at the to-be-adjusted time, calculating the ideal exhaust air volume of the fume hood at the to-be-adjusted time, and inputting the ideal exhaust air volume into a PID control system to control the exhaust air volume of the fume hood. Through the technical scheme of the present application, the accuracy of the control result of the exhaust air volume can be improved.

Description

Technical Field

[0001] This invention relates to the field of exhaust air volume control. In particular, it relates to a method and system for dynamically controlling the exhaust air volume of a fume hood. Background Technology

[0002] Variable air volume (VAV) control technology aims to maintain a constant and safe face velocity at the opening by dynamically adjusting the fan speed or valve opening through real-time monitoring of cabinet door opening, face velocity, or duct static pressure. However, existing control strategies that combine cabinet door position feedforward with face velocity PID closed-loop feedback are prone to temporary drops in face velocity due to lag in air volume adjustment when faced with experimental behavior and thermal disturbances, increasing the risk of pollutant escape. Furthermore, they are not adaptable enough to the actual pollutant generation patterns, changes in thermal buoyancy, and user behavior disturbances.

[0003] Existing technologies use the concentration of harmful gases collected at monitoring points to dynamically control the exhaust air volume of fume hoods in real time. However, they cannot identify the differences in the short-term diffusion intensity of harmful gases between local and peripheral areas caused by the behavior of laboratory personnel. At the same time, they ignore the differences in response lag caused by different gas diffusion delays and disturbance degrees, resulting in low accuracy of exhaust air volume control. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention provides solutions in the following aspects.

[0005] In a first aspect, the present invention provides a dynamic control method for the exhaust air volume of a fume hood, comprising: dividing the fume hood into several analysis layers at equal intervals; taking any preset monitoring point in any analysis layer as a target point; acquiring the characteristic sequence of the target point in history, the characteristic sequence including temperature sequence, humidity sequence and multi-dimensional concentration sequence; acquiring the characteristic sequence of the target point in the period to be detected; taking the last sampling time in the period to be detected as the time to be adjusted; acquiring the exhaust air volume and pressure difference inside the hood at the time to be adjusted; dividing the period to be detected into several sub-segments; taking any sub-segment as a target segment; calculating the disturbance degree of the target point in the target segment based on the temperature sequence, humidity sequence and multi-dimensional concentration sequence; and based on the historical data... A feature sequence is used to train a prediction model and construct a detection window. The length of the detection window is the same as the input length of the prediction model, and the last sampling time in the detection window is the time to be adjusted. The feature sequence of the detection window is input into the prediction model to obtain the prediction window. Any dimension is taken as the target dimension. Based on the concentration sequence of the target dimension in the prediction window, the preset safe concentration value, and the exhaust air volume at the time to be adjusted, the required air volume of the target point at the time to be adjusted is calculated. The required air volume of each monitoring point at the time to be adjusted is obtained by traversing. Based on the required air volume, the disturbance level, and the pressure difference inside the fume hood at the time to be adjusted, the ideal exhaust air volume of the fume hood at the time to be adjusted is calculated. The ideal exhaust air volume is input into the PID control system to control the exhaust air volume of the fume hood.

[0006] Preferably, the step of dividing the time period to be detected into several sub-segments includes: obtaining several candidate schemes, each candidate scheme having a different division result; for a target point in any candidate scheme, calculating the Euclidean distance between the concentration sequences of any two sub-segments in the target dimension, and calculating the first standard deviation of all Euclidean distances; traversing to obtain the first standard deviation of each dimension, and using the sum of all first standard deviations as the stage score of the target point; traversing to obtain the stage score of each monitoring point, and calculating the cumulative stage score of all monitoring points in any candidate scheme; traversing to obtain the cumulative stage score of each candidate scheme, and using the candidate scheme corresponding to the maximum value of the cumulative stage score as the division scheme, and using the division scheme to divide the time period to be detected into several sub-segments.

[0007] Preferably, the calculation of the perturbation degree of the target point in the target segment includes: calculating the absolute value of the first Pearson correlation coefficient between the temperature sequence of the target point in history and the concentration sequence of the target dimension; obtaining the absolute value of the first Pearson correlation coefficient for each monitoring point; and taking the mean of all the absolute values ​​of the first Pearson correlation coefficient as the first correlation degree; obtaining the first correlation degree for each dimension; and calculating the cumulative value of the first correlation degree as the first cumulative value; calculating the absolute value of the second Pearson correlation coefficient between the humidity sequence of the target point in history and the concentration sequence of the target dimension; obtaining the absolute value of the second Pearson correlation coefficient for each monitoring point; and taking the mean of all the absolute values ​​of the second Pearson correlation coefficient as the second correlation degree; obtaining the second correlation degree for each dimension; and calculating the cumulative value of the second correlation degree as the second cumulative value; obtaining the first difference sequence of the concentration sequence of the target point in the target dimension in the target segment; calculating the standard deviation of the difference sequence; calculating the product of the standard deviation and the first correlation degree as the initial product; obtaining the initial product of the target point in each dimension in the target segment; and taking the mean of the first Pearson correlation coefficient for each monitoring point as the second correlation degree; and taking the first Pearson correlation coefficient for each monitoring point as the second cumulative value. The accumulated value of the initial product is used as the first perturbation score of the target point in the target segment; similarly, the product of the difference standard deviation and the second correlation degree is calculated as the first product, and the first product of the target points in each dimension is obtained by traversing through the target segment. The accumulated value of all first products is used as the second perturbation score of the target point in the target segment; the ratio of the first accumulated value to the second accumulated value is calculated as the first ratio, and the second product of the first ratio and the first perturbation score is calculated; the ratio of the second accumulated value to the first accumulated value is calculated as the second ratio, and the second product of the second ratio and the second perturbation score is calculated. The third product of the numbers; the first sum of the second and third products; the first disturbance scores of all monitoring points in the same column as the target point in the fume hood are used to construct the first disturbance sequence, and the standard deviation of the first disturbance sequence is used as the second standard deviation; the second disturbance scores of all monitoring points in the same column as the target point in the fume hood are used to construct the second disturbance sequence, and the standard deviation of the second disturbance sequence is used as the third standard deviation; the second sum of the second and third standard deviations is calculated; the product of the first and second sums is used as the disturbance degree of the target point in the target segment.

[0008] Preferably, the prediction model is a 1D-CNN model or an LSTM model.

[0009] Preferably, the calculation of the required air volume of the target point at the time to be adjusted includes: obtaining the lag order range according to the input length of the prediction model, wherein the number of lag orders in the lag order range is the same as the number of input sampling times; taking the detection point vertically upward from the target point and closest to the target point as the neighboring point, and for the detection window, calculating the autocorrelation coefficient between the concentration sequence of the target point in the target dimension and the concentration sequence of the neighboring point in the target dimension at any lag order for any detection window; traversing to obtain the autocorrelation coefficient at each lag order, corresponding the lag order number one-to-one with the sampling time number in the prediction window, and using the normalized autocorrelation coefficient corresponding to the lag order as the adjustment weight of the sampling time in the prediction window, traversing to obtain the adjustment weight of each sampling time in the target dimension in the prediction window; and comparing the disturbance degree of the target point at the time to be adjusted with all monitoring points. The ratio of the cumulative disturbance levels is used as the third ratio; the first difference between the concentration value in the target dimension at any sampling time in the prediction window and the preset safe concentration value in the target dimension is calculated; the second difference between the concentration value in the target dimension at the time to be adjusted and the preset safe concentration value in the target dimension is calculated; the fourth ratio between the first difference and the second difference is calculated, and the fourth product of the adjustment weight in the target dimension at any sampling time in the prediction window and the fourth ratio is calculated; the fourth product of all sampling times in the target dimension in the prediction window is obtained through iteration, and the fourth product of the fourth product is calculated; the fourth product of all sampling times in the prediction window in each dimension is obtained through iteration, and the cumulative value of all fourth product of the fourth product is used as the comprehensive sum; the product of the third ratio at the time to be adjusted, the exhaust air volume at the time to be adjusted, and the comprehensive sum is used as the required air volume at the target point at the time to be adjusted.

[0010] Preferably, the calculation of the required air volume of the target point at the time to be adjusted further includes: calculating the absolute value of the first Pearson correlation coefficient between the temperature sequence and the concentration sequence of the target dimension in the historical data of the target point, obtaining the absolute value of the first Pearson correlation coefficient for each monitoring point, and taking the mean of all the absolute values ​​of the first Pearson correlation coefficient as the first correlation degree; calculating the absolute value of the second Pearson correlation coefficient between the humidity sequence and the concentration sequence of the target dimension in the historical data of the target point, obtaining the absolute value of the second Pearson correlation coefficient for each monitoring point, and taking the mean of all the absolute values ​​of the second Pearson correlation coefficient as the second correlation degree; calculating the third sum of the first correlation degree and the second correlation degree of the target dimension, obtaining the third sum of each dimension, and calculating the ratio of the third sum of the target dimension to the sum of the third sums of all dimensions as the fifth ratio; and calculating the concentration sequence of the target dimension in the detection window. The mean of the first-order difference sequence difference is used as the first concentration change rate of the target dimension. Similarly, the mean of the first-order difference sequence difference of the concentration sequence of the target dimension in the prediction window is used as the second concentration change rate of the target dimension. The exponential value of the difference between the second concentration change rate and the first concentration change rate is calculated. The third difference between the mean concentration value of the target dimension at all sampling times in the prediction window and the preset safe concentration value of the target dimension is calculated. The fourth difference between the mean concentration value of the target dimension at all sampling times in the detection window and the preset safe concentration value of the target dimension is calculated. The sixth ratio between the third difference and the fourth difference is calculated. The product of the fifth ratio, the exponential value and the sixth ratio is calculated as the fifth product. The fifth product of each dimension is obtained by traversing. The fifth product of all fifth products is accumulated and summed. The product of the exhaust air volume at the time to be adjusted and the fifth product is used as the required air volume of the target point at the time to be adjusted.

[0011] Preferably, the calculation of the ideal exhaust air volume of the fume hood at the time to be adjusted includes: calculating the average disturbance level of all monitoring points in the analysis layer where the target point is located; similarly obtaining the average disturbance level of each analysis layer in the fume hood; calculating the cumulative value of the average disturbance level of all analysis layers; and calculating the seventh ratio of the average disturbance level of the analysis layer where the target point is located to the cumulative value of the average disturbance level; calculating the average demand air volume of the analysis layer where the target point is located; calculating the sixth product of the seventh ratio and the average demand air volume; iterating through each analysis layer to obtain the sixth product and calculating the cumulative value of the sixth product; calculating the eighth ratio of the preset standard pressure difference to the pressure difference inside the hood at the time to be adjusted; and using the product of the eighth ratio and the cumulative value of the sixth product as the ideal exhaust air volume at the time to be adjusted.

[0012] Secondly, the present invention also provides a dynamic control system for the exhaust air volume of a fume hood, comprising: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned dynamic control method for the exhaust air volume of a fume hood is implemented.

[0013] The present invention has the following effects: This invention quantifies the difference in short-term diffusion intensity of harmful gases in local and peripheral areas caused by the behavior of experimental personnel, and combines the nonlinear coupling effect of temperature and humidity changes in the experimental area and other areas to accurately calculate the disturbance level of monitoring points. Based on this disturbance level, it realizes regional adaptive exhaust demand analysis, dynamically improves the air volume response priority of high disturbance areas, effectively avoids experimental accidents caused by the traditional smoothing strategy underestimating the rapid diffusion caused by experimental behavior, and ensures the sensitivity and safety of exhaust control.

[0014] This invention accurately quantifies the spatiotemporal delay characteristics and disturbance differences of harmful gases diffusing from the source to the monitoring point by calculating the correlation coefficient between the target point and its vertically adjacent points at different lag orders. This data is then normalized to adjust the weights and correct the reliability of future concentration predictions. Based on this, the proportion of disturbance at each monitoring point is integrated as a spatial risk amplification coefficient. The deviation between the predicted concentration and the safety threshold is weighted and integrated to calculate the precise required airflow for the area covered by each monitoring point. This scheme, by combining time-dimensional lag compensation with spatial-dimensional disturbance identification, significantly improves the responsiveness and foresight of exhaust airflow control, effectively preventing the escape of harmful gases and experimental safety accidents caused by diffusion delays or control lags. Attached Figure Description

[0015] Figure 1 This is a flowchart of a dynamic control method for exhaust air volume of a fume hood according to an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0017] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0018] The specific scenario addressed by this invention is the intelligent control of exhaust air volume in fume hoods: During experimental operations, the actions of the experimenters, such as arm movements and the handling of equipment, will dynamically disturb the airflow field inside the fume hood, breaking the original stable flow state. This will cause non-uniform differences in the diffusion speed and migration path of harmful gases at different heights and in different spatial areas within the hood. This disturbance will accelerate the turbulent diffusion and short-range migration of harmful gases in local areas (especially near the operating port or in airflow dead zones). If the exhaust system cannot identify and respond to such transient diffusion characteristics induced by human behavior in a timely manner, it is very easy to cause an instantaneous imbalance in the face wind speed, allowing harmful gases to break through the airflow barrier and escape into the operating environment.

[0019] Reference Figure 1 A method for dynamically controlling the exhaust air volume of a fume hood includes steps S1-S4, as detailed below: S1: The fume hood is divided into several analysis layers at equal intervals. Any monitoring point in any analysis layer is taken as the target point. The characteristic sequence of the target point in history is obtained. The characteristic sequence includes temperature sequence, humidity sequence and multi-dimensional concentration sequence. The characteristic sequence of the target point in the period to be detected is obtained. The last sampling time in the period to be detected is taken as the time to be adjusted. The exhaust air volume and pressure difference in the cabinet at the time to be adjusted are obtained.

[0020] In one embodiment, the fume hood is divided into several analysis layers at equal intervals along its height. Multiple monitoring points are arranged at equal intervals along the horizontal direction within each analysis layer. Any pre-set monitoring point in any analysis layer is used as a target point. The multi-dimensional feature sequence formed by the target point during historical experiments is acquired (including temperature sequences collected by temperature sensors, humidity sequences collected by humidity sensors, and multi-dimensional concentration sequences covering various harmful gases such as CO, H2S, Cl2, and NO2 collected by gas sensors). Simultaneously, the feature sequence of the target point during the detection period is acquired. The last sampling time in the detection period is determined as the adjustment time. Combining the preset sliding door width parameters and the safe concentration threshold of harmful gases in the control system, the exhaust air volume sequence is dynamically calculated based on the opening degree, average surface wind speed, and width. Simultaneously, the differential pressure sequence inside the hood is collected in real time by a differential pressure sensor, thereby accurately obtaining the instantaneous exhaust air volume and differential pressure inside the hood corresponding to the adjustment time.

[0021] S2: Divide the time period to be detected into several sub-segments, take any sub-segment as the target segment, and calculate the perturbation degree of the target point in the target segment based on the temperature sequence, humidity sequence and multi-dimensional concentration sequence.

[0022] It should be noted that the diffusion of harmful gases in the flow field inside the fume hood is not a uniform steady-state process, but a nonlinear dynamic behavior influenced by multiple factors such as thermal buoyancy, experimental disturbances, and airflow organization within the hood. This results in significant uncertainty in the path and time of pollutant migration from the source to the monitoring point. Existing control strategies typically reflect the pollution status within the hood in real time based on the concentration at the monitoring point, using a fixed sampling period and a single threshold for feedback adjustment. These strategies fail to consider the instantaneous acceleration effect of local turbulence on the diffusion rate caused by different experimental behaviors (such as rapid arm movement or reagent pouring), nor do they distinguish the response lag gradient caused by differences in the spatial location of the monitoring points (such as fast response near the source and slow response at the edge). Furthermore, they lack a compensation mechanism for the dynamic evolution of gas density and diffusion coefficient caused by changes in temperature and humidity, making it difficult to achieve precise airflow matching within the critical window for pollutant escape.

[0023] In one embodiment, several candidate schemes are randomly selected, and the partitioning results of each candidate scheme are different. For the target point in any candidate scheme, the Euclidean distance between the concentration sequences of any two sub-segments in the target dimension is calculated, and the first standard deviation of all Euclidean distances is calculated. The first standard deviation of each dimension is obtained through traversal, and the sum of all first standard deviations is used as the stage score of the target point. The stage score of each monitoring point is obtained through traversal, and the cumulative value of the stage scores of all monitoring points in any candidate scheme is calculated. The cumulative value of the stage scores of each candidate scheme is obtained through traversal, and the candidate scheme corresponding to the maximum value of the cumulative stage scores is used as the partitioning scheme. The partitioning scheme is used to divide the period to be detected into several sub-segments.

[0024] Take any segment as the target segment, calculate the absolute value of the first Pearson correlation coefficient between the temperature sequence of the target point in history and the concentration sequence of the target dimension, iterate through the data to obtain the absolute value of the first Pearson correlation coefficient for each monitoring point, and take the mean of all the absolute values ​​of the first Pearson correlation coefficient as the first degree of correlation; iterate through the data to obtain the first degree of correlation for each dimension, and calculate the cumulative value of the first degree of correlation as the first cumulative value.

[0025] Calculate the absolute value of the second Pearson correlation coefficient between the humidity sequence of the target point in history and the concentration sequence of the target dimension. Iterate through the data to obtain the absolute value of the second Pearson correlation coefficient for each monitoring point. Take the mean of all the absolute values ​​of the second Pearson correlation coefficient as the second correlation degree. Iterate through the data to obtain the second correlation degree for each dimension. Calculate the cumulative value of the second correlation degree as the second cumulative value.

[0026] Obtain the first-order difference sequence of the concentration sequence of the target point in the target segment in the target dimension, and calculate the difference standard deviation of the first-order difference sequence. Calculate the product of the difference standard deviation and the first correlation degree as the initial product. Iterate through the target points in the target segment to obtain the initial product in each dimension, and accumulate all the initial products as the first perturbation score of the target point in the target segment. Similarly, calculate the product of the difference standard deviation and the second correlation degree as the first product, iterate through the target points in the target segment to obtain the first product in each dimension, and accumulate all the first products as the second perturbation score of the target point in the target segment.

[0027] It should be noted that in the dynamic control scenario of fume hood exhaust air volume, the actions of the experimenters, such as arm movements and the handling of equipment, can cause transient disturbances to the airflow field inside the hood. This disrupts the original stable flow state and accelerates the turbulent diffusion and short-range migration of harmful gases in local areas (especially near the operating port or in airflow dead zones). If the exhaust system cannot identify such non-uniform diffusion characteristics induced by human behavior in a timely manner, it can easily cause instantaneous imbalance of face wind speed, leading to the escape of harmful gases. Existing technologies typically assign weights to monitoring points by analyzing the proportion of abnormal monitoring points and the concentration of harmful gases in multiple monitoring locations to obtain the exhaust air volume. This method relies on the assumption of differences in harmful gas diffusion between experimental and non-experimental locations and the spatial uniformity of temperature and humidity. However, it ignores the significant differences in short-term diffusion intensity between local and edge areas caused by the experimenters' actions, as well as the nonlinear coupling effect between the experimental area and other areas in terms of the degree of temperature and humidity change. This results in systematic bias in the acquired feature data, which in turn underestimates the actual required air volume of the fume hood and increases the risk of experimental accidents caused by insufficient exhaust.

[0028] To this end, this invention deploys monitoring points at equal intervals in layers according to the exhaust direction of the fume hood. First, it uses the feature dataset of historical experiments to calculate the absolute values ​​of Pearson correlation coefficients between temperature, humidity and concentration sequences in each dimension, quantifying the influence weight of temperature and humidity on the diffusion of harmful gases. Then, it combines the first-order difference sequence of concentration characteristics in the current detection period and its difference standard deviation to analyze the instantaneous intensity of disturbance at the monitoring points to obtain the first disturbance score and the second disturbance score. At the same time, it integrates the consistency standard deviation of the disturbance scores of all monitoring points in the exhaust direction (vertical column) of the fume hood to characterize the spatial propagation stability of local disturbances. Finally, it obtains the disturbance degree of the monitoring points through weighted fusion, realizing accurate analysis of exhaust demand based on regional adaptation, effectively avoiding the accident of harmful gas escape caused by the underestimation of rapid diffusion caused by experimental behavior by traditional smoothing strategies.

[0029] Calculate the ratio of the first accumulated value to the second accumulated value as the first ratio, and calculate the second product of the first ratio and the first disturbance score; calculate the ratio of the second accumulated value to the first accumulated value as the second ratio, and calculate the third product of the second ratio and the second disturbance score; calculate the first sum of the second and third products. Construct a first disturbance sequence from the first disturbance scores of all monitoring points in the same column as the target point in the fume hood, and calculate the standard deviation of the first disturbance sequence as the second standard deviation; construct a second disturbance sequence from the second disturbance scores of all monitoring points in the same column as the target point in the fume hood, and calculate the standard deviation of the second disturbance sequence as the third standard deviation; calculate the second sum of the second and third standard deviations; and use the product of the first and second sums as the disturbance degree of the target point in the target segment.

[0030] The construction logic of the disturbance level: The physical meaning of the first and second correlation levels lies in quantifying the linear coupling influence weight of environmental parameters on the diffusion behavior of harmful gases. That is, the first correlation level reflects the driving strength of temperature field changes on pollutant migration rate and diffusion path, and the second correlation level characterizes the modulation effect of humidity fluctuations on gas density, diffusion coefficient, and sedimentation characteristics. The first and second disturbance scores are instantaneous disturbance quantification indicators constructed under the temperature and humidity influence structures, respectively, by combining the standard deviation of the first difference of the concentration sequence in the target segment with the weighted product of the corresponding correlation level. They are used to characterize the dynamic response intensity of the monitoring point under specific environmental coupling conditions affected by experimental behavior. Furthermore, the second and third standard deviations are calculated from the first and second disturbance score sequences of all monitoring points in the exhaust direction (vertical column) of the fume hood, respectively. Their core function is to evaluate the propagation consistency and stability of local disturbances in the vertical spatial dimension. Due to personnel operations during the experiment (such as hand movements), The disturbances caused by arm movement and reagent pouring essentially exhibit localized turbulent diffusion characteristics rather than uniform vertical linear propagation. When the standard deviation is small, it indicates that the disturbance responses of each monitoring point within the same vertical column are highly synchronized, and gas diffusion is dominated by the environment with a relatively stable flow field. Conversely, when the standard deviation is large, it indicates that the gas diffusion velocity around the monitoring point is significantly affected by non-environmental factors (such as human transient disturbances and local vortices), and the diffusion behavior exhibits spatial heterogeneity and non-steady-state characteristics. In this case, it is necessary to improve the response priority of the exhaust air volume in this area through a weighted fusion mechanism of disturbance degree, so as to achieve accurate identification and adaptive control of local rapid diffusion risks.

[0031] S3: Train the prediction model based on the feature sequences in history, construct a detection window. The length of the detection window is the same as the input length of the prediction model, and the last sampling time in the detection window is the time to be adjusted. Input the feature sequence of the detection window into the prediction model to obtain the prediction window. Take any dimension as the target dimension. Calculate the required air volume of the target point in the time to be adjusted based on the concentration sequence of the target dimension in the prediction window, the preset safe concentration value, and the exhaust air volume of the time to be adjusted.

[0032] In one embodiment, a prediction model is trained based on historical feature sequences. The prediction model is either a 1D-CNN (1D Convolutional Neural Network) model or an LSTM (Long Short-Term Memory) model. The historical feature sequences are segmented using a sliding method with the input length of the prediction model to obtain several training samples. For each training sample, a detection window is obtained from these samples. The length of the detection window is the same as the input length of the prediction model, and the last sampling time in the detection window is the time to be adjusted. The output of the prediction model is a prediction window for future consecutive sampling times with the same length as the detection window. The loss function of the prediction model is the mean squared error loss function.

[0033] For example, the prediction model is a 1D-CNN model with an input layer size of 20 x 5, an output size of 20 x 32 for the first convolutional layer using ReLU as the activation function, an output size of 10 x 32 for the first pooling layer, an output size of 10 x 64 for the second convolutional layer using ReLU as the activation function, an output size of 5 x 128 for the second pooling layer, an output size of 5 x 256 for the third convolutional layer using ReLU as the activation function, and an output size of 20 x 5 for the fully connected layer.

[0034] Using any dimension as the target dimension, the range of lag orders is obtained based on the input length of the prediction model. The number of lag orders in the lag order range is the same as the number of sampling times in the input. The detection point vertically upward from the target point and closest to the target point is taken as the neighboring point. For the detection window, the autocorrelation coefficient between the concentration sequence of the target point in the target dimension and the concentration sequence of the neighboring point in the target dimension at any lag order is calculated. The autocorrelation coefficient at each lag order is obtained by iterating through the data. The lag order index is matched one-to-one with the sampling time index in the prediction window. The normalized autocorrelation coefficient corresponding to the lag order is used as the adjustment weight of the sampling time in the prediction window. The adjustment weight of each sampling time in the target dimension in the prediction window is obtained by iterating through the data.

[0035] For example, if the detection window contains 4 sampling times, then the lag order can be 1, 2, 3, and 4. The autocorrelation coefficient with a lag order of 1, after normalization, becomes the adjustment weight of the first sampling time in the target dimension of the prediction window. Similarly, the autocorrelation coefficient with a lag order of 2, after normalization, becomes the adjustment weight of the second sampling time in the target dimension of the prediction window, the autocorrelation coefficient with a lag order of 3, after normalization, becomes the adjustment weight of the third sampling time in the target dimension of the prediction window, and the autocorrelation coefficient with a lag order of 4, after normalization, becomes the adjustment weight of the fourth sampling time in the target dimension of the prediction window.

[0036] It should be noted that in the dynamic control of the exhaust air volume of the fume hood, given the significant time lag in the diffusion of harmful gases generated at the experimental location to the monitoring point, if only the real-time concentration of harmful gases collected at the monitoring point is used for feedback adjustment, an experimental accident often occurs when the concentration exceeds the standard. Existing technology ignores the difference in response lag caused by the diffusion delay of harmful gases and the difference in the degree of disturbance, resulting in untimely control. To address this, this invention trains a prediction model based on feature datasets from historical experiments, constructs a detection window with the same length as the model input to predict concentration sequences at future times, and calculates the autocorrelation coefficients of the target point and its vertically adjacent points at different lag orders. The lag order is then mapped one-to-one with the prediction window time, and the normalized autocorrelation coefficients are used as adjustment weights for the concentration values ​​at each time point within the prediction window to quantify the impact of diffusion delay on prediction reliability. Simultaneously, the ratio of the disturbance level of a monitoring point to the cumulative disturbance level of all monitoring points is used as an amplification factor to weightedly integrate the difference between the predicted concentration and the safe concentration at future times to calculate the required air volume. This allows for an early response before the actual concentration exceeds the limit, significantly improving the sensitivity of exhaust air volume control and preventing untimely control and experimental safety accidents caused by diffusion lag.

[0037] The ratio of the disturbance level of the target point at the time to be adjusted to the cumulative value of the disturbance levels of all monitoring points is used as the third ratio. The first difference between the concentration value of the target dimension at any sampling time in the prediction window and the preset safe concentration value of the target dimension is calculated. The second difference between the concentration value of the target dimension at the time to be adjusted and the preset safe concentration value of the target dimension is calculated. The fourth ratio between the first difference and the second difference is calculated, and the fourth product of the adjustment weight of the target dimension at any sampling time in the prediction window and the fourth ratio is calculated. The fourth product of all sampling times in the target dimension in the prediction window is obtained through iteration, and the fourth product of the fourth product is calculated. The fourth product of all sampling times in the prediction window is obtained through iteration in each dimension, and the cumulative value of all fourth product of the fourth product is used as the comprehensive sum. The product of the third ratio at the time to be adjusted, the exhaust air volume at the time to be adjusted, and the comprehensive sum is used as the required air volume of the target point at the time to be adjusted.

[0038] It should be explained that the third ratio is obtained by calculating the ratio of the disturbance degree of the target monitoring point to the cumulative value of the disturbance degree of all monitoring points in the analysis layer. Its physical meaning is to quantify the disturbance influence weight of the monitoring point relative to the overall space. Since the local turbulence caused by the behavior of the experimenters can significantly accelerate the short-range migration and non-uniform diffusion of harmful gases, the monitoring point with a greater disturbance degree often corresponds to the vicinity of the pollutant source or the area sensitive to airflow disturbance. The concentration change of the monitoring point has a higher requirement for the timeliness of the response to the exhaust demand. Therefore, by adaptively amplifying the demand air volume of the area through the third ratio, the risk of exhaust lag or under-exhaust caused by insufficient disturbance identification can be effectively avoided.

[0039] The weight adjustment is based on the normalization of the autocorrelation coefficients of the target point and its vertically adjacent monitoring points at different lag orders. Its core function is to characterize the modulating effect of the spatiotemporal delay characteristics of the diffusion of harmful gases from the source to the monitoring point on the prediction reliability. The smaller the lag order and the higher the autocorrelation coefficient, the closer the predicted concentration at that moment is coupled with the current flow field state, the greater the reference value for regulation, and the higher the corresponding weight. Conversely, it indicates that the predicted value is strongly affected by the uncertainty of the diffusion path and its regulation priority needs to be reduced.

[0040] Synergistic effect in demand air volume calculation: The third ratio identifies high disturbance risk areas from a spatial dimension and dynamically allocates air volume response priorities. Adjusting weights quantifies the decay of the reliability of predicted concentration due to diffusion lag from a temporal dimension. Through a weighted fusion mechanism, it achieves a dual accurate response to local rapid diffusion and cross-time period concentration evolution, thereby improving the sensitivity of exhaust control and resource utilization efficiency while ensuring experimental safety.

[0041] In another embodiment, calculating the required air volume of the target point at the time to be adjusted further includes: calculating the third sum of the first correlation degree of the target dimension and the second correlation degree of the target dimension, traversing to obtain the third sum of each dimension, and calculating the ratio of the third sum of the target dimension to the cumulative value of the third sum of all dimensions as the fifth ratio.

[0042] The mean of the first-order difference sequence difference of the concentration sequence in the target dimension within the detection window is calculated as the first concentration change rate of the target dimension. Similarly, the mean of the first-order difference sequence difference of the concentration sequence in the target dimension within the prediction window is obtained as the second concentration change rate of the target dimension. The exponential value of the difference between the second concentration change rate and the first concentration change rate is calculated.

[0043] Calculate the third difference between the mean concentration value in the target dimension at all sampling times in the prediction window and the preset safe concentration value in the target dimension; calculate the fourth difference between the mean concentration value in the target dimension at all sampling times in the detection window and the preset safe concentration value in the target dimension; calculate the sixth ratio between the third difference and the fourth difference.

[0044] Calculate the product of the fifth ratio, the exponent value, and the sixth ratio as the fifth product. Iterate through each dimension to obtain the fifth product. Calculate the cumulative sum of the fifth products. Use the product of the exhaust air volume at the time to be adjusted and the cumulative sum of the fifth products as the required air volume at the target point at the time to be adjusted.

[0045] It should be explained that when the fifth ratio is large, it indicates that the migration rate, diffusion coefficient or sedimentation characteristics of the gas are strongly linearly correlated with changes in temperature and humidity, and its concentration evolution is more easily driven by environmental fluctuations. Therefore, it is necessary to increase the monitoring priority and response sensitivity of the gas concentration changes to avoid the risk of reaction lag or insufficient ventilation caused by environmental coupling effects.

[0046] When the predicted rate of change of concentration is lower than the current rate of change (i.e., diffusion tends to slow down), the index value decays as the difference decreases, and the system can appropriately reduce the required air volume to save energy. Conversely, when the predicted rate of change of concentration is significantly higher than the current rate of change (i.e., diffusion accelerates or there are signs of a sudden change), the index value is nonlinearly amplified, and the system intervenes in advance by strengthening the exhaust control intensity, effectively suppressing the risk of escape caused by a sudden increase in concentration.

[0047] Synergistic effects in demand air volume calculation: The fifth ratio identifies highly coupled risk gases from environmental sensitivity and dynamically allocates and regulates resources; the index value quantifies the acceleration characteristics of concentration changes from trend evolution and realizes feedforward compensation; and the weighted fusion mechanism realizes a dual precise response to environmentally driven diffusion and behavior-induced mutations, thereby improving the adaptability and energy efficiency ratio of exhaust control while ensuring experimental safety.

[0048] S4: Iterate through the monitoring points to obtain the required air volume at each time point to be adjusted. Based on the required air volume, the degree of disturbance and the pressure difference inside the cabinet at the time point to be adjusted, calculate the ideal exhaust air volume of the fume hood at the time point to be adjusted. Input the ideal exhaust air volume into the PID control system to control the exhaust air volume of the fume hood.

[0049] It should be noted that in the dynamic control of the exhaust air volume of the fume hood, since experimental operations (such as arm movement and reagent handling) usually occur within a specific height range, the degree of disturbance caused by experimental behavior at monitoring points in different analytical layers within the fume hood exhibits significant spatial heterogeneity. Using uniform air volume control could easily lead to insufficient exhaust in high-risk areas or energy waste in low-risk areas. Therefore, this invention, based on the principle of spatial adaptation, first calculates the average disturbance level of all monitoring points in the analytical layer containing the target point, and then iterates through each analytical layer in the fume hood to obtain the average and cumulative disturbance level. This is achieved by calculating the average disturbance level of that layer. The seventh ratio of the accumulated value is used to quantify the disturbance weight of the layer relative to the overall space. Then, the seventh ratio is multiplied by the average demand air volume of the analysis layer to obtain the sixth product. The sixth product is accumulated for all analysis layers to obtain the space-weighted basic air volume. Furthermore, a pressure difference compensation mechanism is introduced to calculate the eighth ratio of the preset standard pressure difference to the pressure difference inside the cabinet at the time to be adjusted. The product of the eighth ratio and the accumulated sixth product is used as the ideal exhaust air volume at the time to be adjusted. This achieves precise control of the ideal exhaust air volume based on the integration of layered disturbance weights and dynamic pressure difference compensation, ensuring that an effective airflow barrier can be maintained at different experimental heights.

[0050] In one embodiment, the average disturbance level of all monitoring points in the analysis layer where the target point is located is calculated. Similarly, the average disturbance level of each analysis layer in the fume hood is obtained. The cumulative average disturbance level of all analysis layers is calculated, and the seventh ratio of the average disturbance level of the analysis layer where the target point is located to the cumulative average disturbance level is calculated.

[0051] Calculate the average demand air volume of the analysis layer where the target point is located, and calculate the product of the seventh ratio and the sixth product of the average demand air volume; iterate through each analysis layer to obtain the sixth product, and calculate the accumulated value of the sixth product; calculate the eighth ratio of the preset standard pressure difference to the cabinet pressure difference at the time to be adjusted; use the product of the eighth ratio and the accumulated value of the sixth product as the ideal exhaust air volume at the time to be adjusted.

[0052] The calculated ideal exhaust air volume is input as a set value into the PID (proportion integration differentiation) closed-loop control module of the fume hood exhaust system. Combined with the pressure difference sequence inside the hood, dynamic compensation and calibration are performed to ensure that the face velocity at the fume hood opening remains stable within the safe threshold range under complex working conditions such as disturbances caused by experimental personnel behavior, changes in thermal buoyancy, and multi-gas coupling diffusion.

[0053] The system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a dynamic control method for the exhaust air volume of a fume hood according to the first aspect of the present invention.

[0054] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0055] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for dynamically controlling the exhaust air volume of a fume hood, characterized in that, include: Several analysis layers are obtained by dividing the fume hood into equal-spaced layers. Any preset monitoring point in any analysis layer is taken as the target point. The characteristic sequence of the target point in history is obtained. The characteristic sequence includes temperature sequence, humidity sequence and multi-dimensional concentration sequence. The characteristic sequence of the target point in the period to be detected is also obtained. The last sampling time in the period to be detected is taken as the time to be adjusted. The exhaust air volume and pressure difference in the cabinet at the time to be adjusted are obtained. The time period to be detected is divided into several sub-segments. Any sub-segment is taken as the target segment, and the perturbation degree of the target point in the target segment is calculated based on the temperature sequence, humidity sequence and multi-dimensional concentration sequence. A prediction model is trained based on historical feature sequences. A detection window is constructed with the same length as the input length of the prediction model. The last sampling time in the detection window is the time to be adjusted. The feature sequence of the detection window is input into the prediction model to obtain the prediction window. Any dimension is taken as the target dimension. Based on the concentration sequence of the target dimension in the prediction window, the preset safe concentration value, and the exhaust air volume at the time to be adjusted, the required air volume of the target point at the time to be adjusted is calculated. The required air volume of each monitoring point at the time to be adjusted is obtained by iterating through the data. Based on the required air volume, the degree of disturbance and the pressure difference inside the cabinet at the time to be adjusted, the ideal exhaust air volume of the fume hood at the time to be adjusted is calculated. The ideal exhaust air volume is then input into the PID control system to control the exhaust air volume of the fume hood.

2. The method for dynamic control of exhaust air volume of a fume hood according to claim 1, characterized in that, The process of dividing the time period to be detected into several sub-segments includes: Several candidate solutions are obtained, and the partitioning results of each candidate solution are different; For any target point in a candidate scheme, calculate the Euclidean distance between any two sub-segments in the concentration sequence of the target dimension, and calculate the first standard deviation of all Euclidean distances; iterate through each dimension to obtain the first standard deviation, and use the sum of all first standard deviations as the stage score of the target point; Iterate through each monitoring point to obtain its stage score, and calculate the cumulative stage score of all monitoring points in any candidate scheme. The system iterates through each candidate scheme to obtain the cumulative stage score. The candidate scheme corresponding to the maximum cumulative stage score is used as the partitioning scheme. The partitioning scheme is then used to divide the time period to be detected into several sub-segments.

3. The method for dynamic control of exhaust air volume of a fume hood according to claim 1, characterized in that, The degree of disturbance of the target point in the calculated target segment includes: Calculate the absolute value of the first Pearson correlation coefficient between the temperature sequence of the target point in history and the concentration sequence of the target dimension. Iterate through the data to obtain the absolute value of the first Pearson correlation coefficient for each monitoring point. Take the mean of all the absolute values ​​of the first Pearson correlation coefficient as the first degree of correlation. Iterate through the data to obtain the first degree of correlation for each dimension. Calculate the cumulative value of the first degree of correlation as the first cumulative value. Calculate the absolute value of the second Pearson correlation coefficient between the humidity sequence of the target point in history and the concentration sequence of the target dimension. Iterate through the data to obtain the absolute value of the second Pearson correlation coefficient for each monitoring point. Take the mean of all the absolute values ​​of the second Pearson correlation coefficient as the second correlation degree. Iterate through the data to obtain the second correlation degree for each dimension. Calculate the cumulative value of the second correlation degree as the second cumulative value. Obtain the first-order difference sequence of the concentration sequence of the target point in the target segment in the target dimension, and calculate the difference standard deviation of the first-order difference sequence. Calculate the product of the difference standard deviation and the first correlation degree as the initial product. Iterate through the target points in the target segment to obtain the initial product in each dimension, and accumulate all the initial products as the first perturbation score of the target point in the target segment. Similarly, calculate the product of the difference standard deviation and the second correlation degree as the first product, iterate through the target points in the target segment to obtain the first product in each dimension, and accumulate all the first products as the second perturbation score of the target point in the target segment. Calculate the ratio of the first accumulated value to the second accumulated value as the first ratio, and calculate the second product of the first ratio and the first perturbation fraction; calculate the ratio of the second accumulated value to the first accumulated value as the second ratio, and calculate the third product of the second ratio and the second perturbation fraction; calculate the first sum of the second product and the third product; The first perturbation scores of all monitoring points in the same column as the target point in the fume hood are used to construct a first perturbation sequence, and the standard deviation of the first perturbation sequence is calculated as the second standard deviation; the second perturbation scores of all monitoring points in the same column as the target point in the fume hood are used to construct a second perturbation sequence, and the standard deviation of the second perturbation sequence is calculated as the third standard deviation; the second sum of the second standard deviation and the third standard deviation is calculated. The product of the first sum and the second sum is taken as the degree of perturbation of the target point in the target segment.

4. The method for dynamic control of exhaust air volume of a fume hood according to claim 1, characterized in that, The prediction model is either a 1D-CNN model or an LSTM model.

5. The method for dynamic control of exhaust air volume of a fume hood according to claim 1, characterized in that, The calculation of the required air volume at the target point during the time to be adjusted includes: The lag order range is obtained based on the input length of the prediction model. The number of lag orders in the lag order range is the same as the number of sampling times of the input. The nearest detection point vertically upwards from the target point is taken as the neighboring point. For the detection window, the autocorrelation coefficient between the concentration sequence of the target point in the target dimension and the concentration sequence of the neighboring point in the target dimension at any lag order is calculated. The autocorrelation coefficient at each lag order is obtained by iterating through the detection window. The lag order number is matched one-to-one with the sampling time number in the prediction window. The normalized autocorrelation coefficient corresponding to the lag order is used as the adjustment weight of the sampling time in the prediction window. The adjustment weight of each sampling time in the target dimension in the prediction window is obtained by iterating through the detection window. The ratio of the disturbance level of the target point at the time to be adjusted to the cumulative value of the disturbance levels of all monitoring points is used as the third ratio. Calculate the first difference between the concentration value in the target dimension at any sampling time in the prediction window and the preset safe concentration value in the target dimension; calculate the second difference between the concentration value in the target dimension at the time to be adjusted and the preset safe concentration value in the target dimension; calculate the fourth ratio of the first difference and the second difference, and calculate the fourth product of the adjustment weight in the target dimension at any sampling time in the prediction window and the fourth ratio. Iterate through all sampling times in the prediction window to obtain the fourth product in the target dimension, and calculate the fourth product cumulative sum. Iterate through the sample time in the prediction window to obtain the fourth-squared cumulative sum of each dimension, and use the sum of all the fourth-squared cumulative sums as the comprehensive sum. The product of the third ratio at the time to be adjusted, the exhaust air volume at the time to be adjusted, and the comprehensive sum is taken as the required air volume at the target point at the time to be adjusted.

6. The method for dynamic control of exhaust air volume of a fume hood according to claim 5, characterized in that, The calculation of the required air volume at the target point during the time to be adjusted also includes: Calculate the absolute value of the first Pearson correlation coefficient between the temperature sequence of the target point in history and the concentration sequence of the target dimension. Iterate through each monitoring point to obtain the absolute value of the first Pearson correlation coefficient. The mean of all the absolute values ​​of the first Pearson correlation coefficient is taken as the first degree of correlation. Calculate the absolute value of the second Pearson correlation coefficient between the humidity sequence of the target point in history and the concentration sequence of the target dimension. Iterate through each monitoring point to obtain the absolute value of the second Pearson correlation coefficient. The mean of all the absolute values ​​of the second Pearson correlation coefficient is taken as the second correlation degree. Calculate the third sum of the first relevance of the target dimension and the second relevance of the target dimension, iterate through each dimension to obtain the third sum, and calculate the ratio of the third sum of the target dimension to the sum of the third sums of all dimensions as the fifth ratio. The mean of the first-order difference sequence differences of the concentration sequence in the target dimension within the detection window is calculated as the first concentration change rate of the target dimension. Similarly, the mean of the first-order difference sequence differences of the concentration sequence in the target dimension within the prediction window is obtained as the second concentration change rate of the target dimension. The exponential value of the difference between the second concentration change rate and the first concentration change rate is calculated. Calculate the third difference between the mean concentration value in the target dimension at all sampling times in the prediction window and the preset safe concentration value in the target dimension; calculate the fourth difference between the mean concentration value in the target dimension at all sampling times in the detection window and the preset safe concentration value in the target dimension; calculate the sixth ratio between the third difference and the fourth difference. Calculate the product of the fifth ratio, the exponent value, and the sixth ratio as the fifth product. Iterate through each dimension to obtain the fifth product. Calculate the cumulative sum of the fifth products. Use the product of the exhaust air volume at the time to be adjusted and the cumulative sum of the fifth products as the required air volume at the target point at the time to be adjusted.

7. The method for dynamic control of exhaust air volume of a fume hood according to claim 1, characterized in that, The ideal exhaust air volume of the fume hood at the time to be adjusted includes: Calculate the average disturbance level of all monitoring points in the analysis layer where the target point is located. Similarly, obtain the average disturbance level of each analysis layer in the fume hood. Calculate the cumulative average disturbance level of all analysis layers and the seventh ratio of the average disturbance level of the analysis layer where the target point is located to the cumulative average disturbance level. Calculate the average demand air volume of the analysis layer where the target point is located, and calculate the product of the seventh ratio and the sixth product of the average demand air volume; Iterate through each analysis layer to obtain the sixth product, and calculate the sixth product accumulation value; Calculate the eighth ratio of the preset standard pressure difference to the pressure difference inside the cabinet at the time to be adjusted; The product of the eighth ratio and the sixth accumulated value is taken as the ideal exhaust air volume at the time to be adjusted.

8. A dynamic control system for the exhaust air volume of a fume hood, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a dynamic control method for exhaust air volume of a fume hood according to any one of claims 1-7.

Images