Ventilated cabinet-based air volume control valve control method
By combining various environmental characteristics and a random forest model to predict wind speed errors, the opening of the airflow control valve of the fume hood is adjusted, thus solving the problem of wind speed fluctuation on the fume hood surface and achieving stable control in dynamic environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 浙江科恩实验设备股份有限公司
- Filing Date
- 2025-08-22
- Publication Date
- 2026-08-04
AI Technical Summary
Existing fume hoods have a single airflow control strategy, which cannot quickly adapt to dynamically changing operating environments, resulting in large fluctuations in face velocity, or even deviations from the safe range.
By acquiring the opening degree of the air volume control valve of the fume hood in real time, and combining various environmental characteristics such as door opening degree, air density and wind pressure difference, the random forest model is used to predict the wind speed error, and the opening degree of the air volume control valve is adjusted to stabilize the surface wind speed.
Under conditions of frequent cabinet door operation and fluctuations in temperature, humidity and air pressure, it can effectively maintain the face velocity within a safe range, improving the accuracy and stability of airflow regulation.
Smart Images
Figure CN121139738B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ventilation equipment control. More particularly, it relates to a method for controlling the airflow of a fume hood-based airflow control valve. Background Technology
[0002] In laboratory settings, fume hoods are critical equipment for ensuring a safe experimental environment. They maintain a stable face velocity by regulating airflow to prevent the spread of harmful gases. However, during actual use, the face velocity can change due to factors such as the opening and closing of the hood door. Too low a face velocity may lead to the leakage of harmful gases or particulate matter, while too high a face velocity may affect the stability of experimental operations. Therefore, precise control of airflow is necessary to keep the face velocity stable within a safe range.
[0003] Chinese patent application CN115971194A discloses a method and apparatus for controlling variable air volume (VAV) in a fume hood, comprising: acquiring the face velocity of the fume hood; in response to determining that the face velocity does not meet a preset requirement, calculating the required exhaust volume of the fume hood based on the opening degree of the fume hood window displacement sensor and the set face velocity value; and controlling the air volume of the VAV butterfly valve based on the exhaust volume value.
[0004] However, the operating environment and usage conditions of fume hoods are dynamic. Controlling the air volume of variable air volume butterfly valves solely by adjusting the opening of the fume hood window displacement sensor and setting the face velocity value cannot quickly adapt to these changes, resulting in untimely air volume adjustment, large fluctuations in face velocity, and even deviations from the safe range. Summary of the Invention
[0005] To address the technical problem of the aforementioned fume hoods having a single control strategy for variable air volume butterfly valves, resulting in untimely air volume adjustment, large fluctuations in face velocity, and even deviations from the safe range, this invention provides a control method for air volume control valves based on fume hoods.
[0006] Airflow control valve control methods based on fume hoods include: The opening degree of the air volume control valve of the fume hood is acquired in real time, and the opening degree of the air volume control valve is adjusted in real time based on the environmental characteristics of the fume hood to keep the real-time face velocity in the fume hood stable within a safe range. The adjustment process includes: collecting data from a preset time period prior to the current moment and preprocessing it to obtain historical sample data, which includes environmental feature data and wind speed error data. The wind speed error data is the difference between the measured surface wind speed and the target surface wind speed at the time of collection. The comprehensive correlation value between each type of environmental feature data and the wind speed error data is calculated. A random forest model is trained based on the historical sample data and the corresponding comprehensive correlation value, and the weight of each decision tree is calculated during the model training process. Real-time environmental feature data is input into the trained model, and the predicted wind speed error of a single tree is output. Then, the predicted wind speed error is obtained by weighted summation. The opening of the real-time airflow control valve is adjusted according to the final predicted wind speed error value.
[0007] This invention comprehensively considers various environmental characteristics and quantifies the impact of environmental characteristics such as cabinet door opening, air density, and wind pressure difference on wind speed error by calculating static and dynamic correlation values. It fully reflects the relationship between environmental characteristics and face wind speed, and solves the problem of insufficient control accuracy caused by neglecting complex environmental correlations in traditional methods. It can still maintain face wind speed within a safe range under conditions such as frequent cabinet door operation and fluctuations in temperature, humidity, and air pressure.
[0008] Preferably, the environmental characteristic data includes cabinet door opening data, air density data, and air pressure difference data inside and outside the cabinet.
[0009] Preferably, the calculation process of the comprehensive correlation value is as follows: calculate the static correlation value and dynamic correlation value between each type of environmental feature data and wind speed error data, and sum the corresponding static correlation value and dynamic correlation value by weight to obtain the comprehensive correlation value between each type of environmental feature data and wind speed error data.
[0010] Preferably, the calculation process for the static correlation value is as follows: For each type of environmental feature data, the environmental feature data and wind speed error data in all historical sample data are discretized, and the discretized intervals are assigned discrete symbols. The joint probability distribution and marginal probability distribution of the discrete symbols of the environmental feature data and the discrete symbols of the wind speed error data are calculated. Based on the joint probability distribution and marginal probability distribution, the mutual information value of the environmental feature data and the wind speed error data is calculated to obtain the static correlation value.
[0011] Discretization transforms continuous data into discrete symbols, simplifying the data processing and improving computational efficiency. Furthermore, discretization reduces the impact of noise in the data, enhancing its stability and reliability. By quantifying the static correlation between environmental characteristic data and wind speed error data through mutual information values, it intuitively reflects the degree of influence of each environmental characteristic data on wind speed error data.
[0012] Preferably, the calculation process for the static correlation value is as follows: For each type of environmental feature data, the number of that environmental feature in all historical sample data is binned, and the wind speed error data in the historical sample data of the environmental feature data in each bin is counted; the variance of the wind speed error within each bin and the range of the wind speed error between all bins are calculated; the ratio of the weighted sum of the variances of the wind speed error within each bin to the range of the wind speed error between each bin is used as the static correlation value.
[0013] Preferably, the calculation process for the dynamic correlation value is as follows: All historical sample data are arranged in chronological order of collection to form a historical sample sequence. For each type of environmental characteristic data, the absolute value of the Pearson correlation coefficient between the environmental characteristic data and the wind speed error data within the collection period is used as the dynamic correlation value.
[0014] By calculating the Pearson correlation coefficient, we can capture the dynamic relationship between environmental characteristic data and wind speed error data, reflecting their temporal correlation. The larger the value, the stronger the linear correlation between the two time series.
[0015] Preferably, the calculation process for the dynamic correlation value is as follows: All historical sample data are arranged in the order of collection time to form a historical sample sequence. For each type of environmental feature data, state sample data is constructed based on the collection time. The state sample data includes environmental feature vector, historical wind speed error vector, and future wind speed error value. The transfer entropy of environmental feature data to wind speed error data is calculated based on the state sample data to obtain the dynamic correlation value.
[0016] Preferably, the calculation process for the transfer entropy is as follows: The equal-frequency binning method is used to divide the environmental feature vector, historical wind speed error vector, and future wind speed error value of all state sample data into multiple intervals, and a discrete symbol is assigned to each interval. Based on the discrete symbol, the joint probability, the first conditional probability, and the second conditional probability of the environmental feature vector, the historical wind speed error vector, and the future wind speed error value are calculated. The transfer entropy of the environmental feature data to the wind speed error data is calculated based on the joint probability, the first conditional probability, and the second conditional probability.
[0017] Transfer entropy can quantify the dynamic causal impact of environmental characteristic data on wind speed error data, revealing how changes in environmental characteristics affect future wind speed errors. It reflects the actual dynamic relationship better than simple correlation analysis. Moreover, the calculation of transfer entropy is based on probability distribution, which has a certain robustness to outliers and noise, thereby improving the stability and reliability of the entire system.
[0018] Preferably, the calculation process for the weight of each decision tree is as follows: For each decision tree, obtain the comprehensive correlation value between all types of environmental feature data and wind speed error data in its input, and take the average of all comprehensive correlation values as the matching degree of the decision tree; the ratio of this matching degree to the sum of the matching degrees of all decision trees is taken as the weight of the decision tree.
[0019] The random selection of environmental feature data types and historical sample data by each decision tree increases the randomness of the model and reduces the risk of overfitting. Moreover, the use of different combinations of environmental feature data types by different trees increases the diversity of the model and improves the generalization ability of the overall model. By calculating the matching degree of each decision tree, the contribution of different combinations of environmental feature data to the model can be evaluated. By normalizing the matching degree as a weight, it can be ensured that the contribution of each decision tree is proportional to its matching degree, thereby improving the interpretability of the model.
[0020] Preferably, the process of adjusting the real-time airflow control valve opening based on the final wind speed error prediction value is as follows: set a fixed proportional coefficient, calculate the product of the opposite of the final wind speed error prediction value and the fixed proportional coefficient, calculate the sum of the product and the real-time airflow control valve opening to obtain the target opening, and adjust the real-time airflow control valve opening according to the target opening.
[0021] The present invention has the following effects: 1. This invention comprehensively considers various environmental characteristics and quantifies the impact of environmental characteristics such as cabinet door opening, air density, and wind pressure difference on wind speed error by calculating static and dynamic correlation values. It fully reflects the relationship between environmental characteristics and face wind speed, and solves the problem of insufficient control accuracy caused by neglecting complex environmental correlations in traditional methods. It can still maintain face wind speed within a safe range under conditions such as frequent cabinet door operation and fluctuations in temperature, humidity, and air pressure.
[0022] 2. This invention focuses on key influencing factors by combining environmental feature importance quantification methods such as mutual information and transfer entropy. It uses random forest multi-tree weighted fusion prediction, which has higher prediction accuracy and stronger anti-overfitting ability compared with a single model. It provides a reliable leading error reference for wind volume regulation and reduces the amplitude of wind speed fluctuations. Attached Figure Description
[0023] Figure 1 This is a flowchart of steps S1-S4 in a method for controlling the air volume of a fume hood based on an embodiment of the present invention. Detailed Implementation
[0024] 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.
[0025] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0026] This invention takes into account that the operating environment and usage conditions of fume hoods are dynamic. If the opening of the air volume control valve is adjusted by a single control strategy, the actual face velocity will fluctuate frequently and cannot be stabilized within a safe range. Therefore, this invention proposes a method that takes into account multiple environmental factors and accurately controls the air volume control valve of the fume hood.
[0027] In the embodiments of the invention, the method for controlling the airflow control valve of a fume hood involves acquiring the opening degree of the airflow control valve of the fume hood in real time and adjusting the opening degree of the airflow control valve based on the environmental characteristics of the fume hood to stabilize the real-time face velocity inside the fume hood within a safe range. The adjustment process includes steps S1 to S4, as described above. Figure 1 Specifically as follows: S1. Collect data within a preset time period before the current moment and preprocess it to obtain historical sample data. The historical sample data includes environmental characteristic data and wind speed error data. The wind speed error data is the difference between the measured surface wind speed and the target surface wind speed at the time of collection.
[0028] Real-time monitoring of fume hood door opening data using displacement sensors. The unit is a percentage, for example, 0% means completely off, and 100% means completely on.
[0029] Obtain air density data , In the formula, Indicates air density; This indicates the total air pressure, obtained from a pressure sensor. This indicates the partial pressure of water vapor, and the unit is Pa (Pascal). , Relative humidity is indicated by a temperature and humidity sensor. The saturated water vapor pressure is expressed through thermodynamic temperature. Obtained by looking up the table. This represents the water vapor gas constant.
[0030] The air pressure difference between the inside and outside of the cabinet is measured using a barometric pressure sensor. The unit is Pa (Pascal).
[0031] Environmental characteristic data includes cabinet door opening data, air density data, and air pressure difference data between the inside and outside of the cabinet. Environmental characteristic data and face wind speed data from 50-100 seconds prior to the current moment are collected. The collected data undergoes a moving median filter to eliminate sudden noise. If a data point deviates from the mean by more than three times the standard deviation, the mean of the preceding and following values is used instead. Z-score standardization is applied to the cabinet door opening data, air density data, and air pressure difference data between the inside and outside of the cabinet.
[0032] The target wind speed was set at 0.65 m / s. According to experimental safety standards, the face velocity of a fume hood is typically 0.5 m / s to 0.8 m / s, and the target wind speed can be adjusted within this range. The difference between the preprocessed face velocity data and the target wind speed was calculated to obtain the wind speed error data. The opening degree data, air density data, pressure difference data inside and outside the hood, and wind speed error data corresponding to a single acquisition moment constituted historical sample data. All historical sample data were arranged in the order of acquisition time to form a historical sample sequence.
[0033] S2. Calculate the comprehensive correlation value between each type of environmental characteristic data and wind speed error data.
[0034] The calculation process of the comprehensive correlation value is as follows: calculate the static correlation value and dynamic correlation value between each type of environmental characteristic data and wind speed error data, and then sum the corresponding static correlation value and dynamic correlation value by weight to obtain the comprehensive correlation value between each type of environmental characteristic data and wind speed error data.
[0035] By comprehensively considering various environmental characteristics, such as cabinet door opening, air density, and air pressure difference between the inside and outside of the cabinet, and calculating the static and dynamic correlation values between each environmental characteristic data and wind speed error data, the relationship between each environmental characteristic and face wind speed is fully reflected, providing reliable data for subsequent adjustment of the air volume control valve.
[0036] The calculation process of the static correlation value is as follows: For each type of environmental feature data, the environmental feature data and wind speed error data in all historical sample data are discretized, and the discretized intervals are assigned discrete signs respectively; the joint probability distribution and marginal probability distribution of the discrete signs of the environmental feature data and the discrete signs of the wind speed error data are calculated; based on the joint probability distribution and marginal probability distribution, the mutual information value of the environmental feature data and the wind speed error data is calculated to obtain the static correlation value.
[0037] The calculation process for the static correlation value between the three types of environmental characteristic data and wind speed error data is the same. To avoid repetitive description, this embodiment will use the calculation of the static correlation value between cabinet door opening data and wind speed error data as an example for explanation.
[0038] The cabinet door opening data and wind speed error data in all historical sample data were divided into 10-20 intervals using the equidistant division method. The specific number of intervals can be adjusted according to the data distribution. For example, the cabinet door opening data can be divided into the following intervals according to the percentage of opening: [0%-10%], [11%-20%], ..., [91%-100%], and the wind speed error data can be divided into the following intervals according to the actual fluctuation range of wind speed error: [-0.7, -0.6), [-0.6, -0.5), ..., [0.6, 0.7].
[0039] For each interval of cabinet door opening data, each interval corresponds to a unique discrete symbol, such as... , This represents the total number of intervals; for each interval of wind speed error data, each interval also corresponds to a unique discrete symbol, such as... , This represents the total number of intervals. For example, when the cabinet door opening data is 35%, it falls within the interval [31%-40%], corresponding to the discrete symbol 4; when the wind speed error data is -0.55m / s, it falls within the interval [-0.6, -0.5), corresponding to the discrete symbol 2.
[0040] Calculate the joint probability distribution of cabinet door opening data and wind speed error data. Assume the total number of historical sample data is N, and the th... The interval and the first wind speed error data The number of times each interval appears together is Then the formula for calculating the joint probability is: In the formula, This represents the range of cabinet door opening data. Indicates the first Each interval This represents the range of wind speed error data. Indicates the first A range.
[0041] If the cabinet door opening data falls within the range of [31%-40%] (corresponding to discrete symbol 4), and the wind speed error data falls within the range of [-0.6, -0.5) (corresponding to discrete symbol 2) 200 times, and the total sample size N is 10000, then... .
[0042] Calculate the marginal probability distribution of the cabinet door opening data. Statistically analyze the first... The number of times each interval appears in all historical sample data Then the marginal probability distribution If the cabinet door opening data appears 100 times within the range of [31%-40%] (corresponding to discrete symbol 4), then... .
[0043] Calculate the marginal probability distribution of wind speed error data. Statistically analyze the first... The number of times each interval appears in all historical sample data Then the marginal probability distribution .
[0044] Substituting the joint probability distribution and marginal probability distribution into the mutual information formula, the mutual information value between the cabinet door opening data and the wind speed error data is calculated. The specific calculation formula is as follows: In the formula, This represents the environmental characteristic of cabinet door opening data; This indicates wind speed error data; This represents the mutual information value between cabinet door opening data and wind speed error data; This represents the total number of intervals for cabinet door opening data; The first data representing the cabinet door opening degree A range; This represents the total number of intervals for wind speed error data. This represents the range of cabinet door opening data; This represents the range of wind speed error data; The first data representing the cabinet door opening degree The interval and the first wind speed error data The joint probability distribution of the intervals; The first data representing the cabinet door opening degree Marginal probability distribution of each interval; The first part representing the wind speed error data Marginal probability distribution of each interval; It represents the logarithm to the base 2.
[0045] A higher mutual information value indicates a stronger static correlation between the cabinet door opening data and the wind speed error data, meaning the impact of the cabinet door opening data on the wind speed error data is more significant. This quantitative method can intuitively reflect the interdependence between two variables, providing a basis for subsequent analysis and decision-making.
[0046] The mutual information value is used as the static correlation value between the cabinet door opening data and the wind speed error data, as specifically represented below: In another embodiment, the static correlation value is calculated as follows: For each type of environmental feature data, the number of that environmental feature in all historical sample data is binned, and the wind speed error data in the historical sample data where the environmental feature data in each bin is located is counted; the intra-bin wind speed error variance and the inter-bin wind speed error range of all bins are calculated; the ratio of the weighted sum of the intra-bin wind speed error variances to the inter-bin wind speed error range is used as the static correlation value.
[0047] Since the calculation process for the static correlation value between the three types of environmental characteristic data and wind speed error data is the same, to avoid repetitive description, we will still take the calculation of the static correlation value between cabinet door opening data and wind speed error data as an example for explanation.
[0048] Divide all historical sample data on cabinet door opening into n bins, preferably 5-10 bins, using a range-based or equal-frequency partitioning method. For example, divide the cabinet door opening data into 10 bins at 10% intervals: [0%-10%], [11%-20%], ..., [91%-100%]. Each bin should contain a sufficient number of samples (ideally greater than or equal to 30) to avoid biased results due to data sparsity.
[0049] For each compartment, wind speed error data is extracted from the historical sample data corresponding to the compartment door opening data, and the variance of the error inside the compartment is calculated. The specific calculation formula is as follows: In the formula, The first data representing the cabinet door opening degree The variance of the in-box error of each sub-box; Indicates the first The sample size of cabinet door opening data for each compartment; Indicates the first The first one in the first compartment One sample; Indicates the first The first one in the first compartment Wind speed error data corresponding to each sample; Indicates the first The mean of all wind speed error data for each sub-box.
[0050] The variance of the error within a compartment measures the degree of fluctuation in the wind speed error data within that compartment. If the variance of the error within a compartment is very small, it indicates that the wind speed error data within that compartment is relatively concentrated and fluctuates little, suggesting that the wind speed error changes within that compartment are relatively stable, and that the door opening data has a strong interpretive ability for the wind speed error data.
[0051] Calculate the range of errors between all compartments to measure the overall difference in errors between different compartments. The specific calculation formula is as follows: In the formula, This indicates the range of errors between boxes; This represents the maximum value of the average wind speed error within all compartments; This represents the minimum mean wind speed error across all compartments. If an abnormal compartment exists (e.g., sample size less than 30), the mean wind speed error of that compartment is excluded from the calculation.
[0052] The inter-compartment error range represents the difference between the maximum and minimum values of the mean wind speed error within all compartments, measuring the degree of difference in the mean error between different compartments. A large inter-compartment error range indicates a significant difference in the mean wind speed error between different compartments, suggesting that the door opening data can significantly distinguish the wind speed error data within different compartments, meaning that the wind speed error data within different compartments are clearly different.
[0053] The ratio of the weighted sum of the variances of the air velocity errors inside the chamber to the range of the air velocity errors between chambers is used as the static correlation value. The calculation formula is as follows: In the formula, This represents the static correlation value between cabinet door opening data and wind speed error data; This indicates the range of errors between boxes; The first data representing the cabinet door opening degree The variance of the in-box error of each sub-box; The total number of compartments representing the cabinet door opening data; This indicates the total number of historical sample data; Indicates the first The ratio of the sample size of the cabinet door opening data of each compartment to the total sample size of the cabinet door opening data is used to weight and balance the sample size differences between compartments.
[0054] When the static correlation value is larger, the cabinet door opening data has a stronger ability to interpret the grouping of wind speed error data. If the static correlation value between cabinet door opening data and wind speed error data is significantly higher than the static correlation value between other environmental feature data and wind speed error data, it means that the cabinet door opening data performs better in distinguishing wind speed errors and is an important explanatory variable.
[0055] Similarly, the static correlation values between air density data and wind speed error data, and the static correlation values between the air pressure difference data inside and outside the cabinet and wind speed error data can be calculated.
[0056] The calculation process of the dynamic correlation value is as follows: All historical sample data are arranged in the order of collection time to form a historical sample sequence. For each type of environmental characteristic data, the absolute value of the Pearson correlation coefficient between the environmental characteristic data and the wind speed error data within the collection time is used as the dynamic correlation value.
[0057] When calculating the dynamic correlation value between environmental feature data and wind speed error data, the time series of historical sample data needs to be considered. Since the calculation process for the dynamic correlation value between the three types of environmental feature data and wind speed error data is the same, to avoid repetitive description, this embodiment uses the calculation of the dynamic correlation value between cabinet door opening data and wind speed error data as an example for explanation.
[0058] All historical sample data are arranged in chronological order to form a historical sample sequence. The cabinet door opening data sequence and wind speed error data sequence are extracted from the historical sample sequence. Based on the cabinet door opening data sequence and wind speed error data sequence, the Pearson correlation coefficient between the cabinet door opening data and the wind speed error data is calculated. The larger the absolute value of the Pearson correlation coefficient, the stronger the linear correlation between the cabinet door opening data and the wind speed error data, and the higher the importance of the cabinet door opening data.
[0059] The formula for calculating the Pearson correlation coefficient between cabinet door opening data and wind speed error data is as follows: In the formula, The Pearson correlation coefficient represents the relationship between cabinet door opening data and wind speed error data. This represents all data collection times within the data collection period; Indicates the first The numerical value of the cabinet door opening at any given time; This represents the average value of all cabinet door opening data within the data collection period. Indicates the first The numerical value of the wind speed error data at any given time; This represents the average value of all wind speed error data within the data collection period.
[0060] The absolute value of the Pearson correlation coefficient is used as the dynamic correlation value between the cabinet door opening data and the wind speed error data, as shown below: The Pearson correlation coefficient ranges from -1 to 1, with positive and negative values indicating the direction of the correlation and the absolute value indicating the strength of the correlation. Ranking the Pearson correlation coefficients between the three environmental feature data and the wind speed error data, the larger the Pearson correlation coefficient, the stronger the linear correlation between the environmental feature data and the wind speed error data, and the higher the importance of the environmental feature data.
[0061] In another embodiment, the calculation process of the dynamic correlation value is as follows: all historical sample data constitute a historical sample sequence in the order of collection time. For each type of environmental feature data, state sample data is constructed based on the collection time. The state sample data includes environmental feature vector, historical wind speed error vector, and future wind speed error value. The transfer entropy of environmental feature data to wind speed error data is calculated based on the state sample data to obtain the dynamic correlation value.
[0062] Since the calculation process for the dynamic correlation value between the three types of environmental characteristic data and wind speed error data is the same, to avoid repetitive description, we will still take the calculation of the dynamic correlation value between cabinet door opening data and wind speed error data as an example for explanation.
[0063] Based on historical sample sequences, the previous data on cabinet door opening were extracted. Each time value ( As an embedding dimension, this embodiment sets =3), which constitutes the environmental feature vector at time t: Before taking wind speed error data Each time value ( recommend =3), which constitutes the historical wind speed error vector at time t: Take the next moment value of the wind speed error data as the future value of the wind speed error: Environmental feature vectors, historical wind speed error vectors, and future wind speed error values constitute the state sample data, ultimately yielding... Group state sample data. This represents all data collection times within the data collection period. For the embedded dimension.
[0064] Discretize each individual element in the environmental feature vector of all state sample data, that is, discretize each element in the environmental feature vector of all states. Divide into several intervals (each interval corresponds to a discrete symbol). Divide into several intervals (each interval corresponds to a discrete symbol), ... It is divided into several intervals (each interval corresponds to a discrete symbol). When environmental feature vectors are converted into discrete symbols, one of the environmental feature vectors may be represented as... .
[0065] Discretize each element of the wind speed error history vector in all state sample data, following the same process as discretizing the environmental feature vectors described above. After converting the wind speed error history vector into discrete notation, one of the wind speed error history vectors might be represented as... .
[0066] Discretize the future values of wind speed error in all state sample data, and assign a discrete sign to each interval.
[0067] After converting the state sample data into discrete symbols, the probability of the simultaneous occurrence of the environmental feature vector, the historical wind speed error vector, and the future wind speed error value is calculated, i.e., the joint probability. The calculation formula is as follows: in, This represents a sample where the environmental feature vector, historical wind speed error vector, and future wind speed error value appear simultaneously after being converted into discrete symbols. This represents the total amount of state sample data. For example, hour, , , If the number of samples appearing simultaneously is 1, then .
[0068] Calculate the first conditional probability: given the historical wind speed error vector, calculate the probability of the future wind speed error value. In the formula, This represents the joint probability of the historical wind speed error vector and the future wind speed error value. The marginal probability, representing the historical vector of wind speed error, is calculated using existing techniques and will not be elaborated upon here.
[0069] Calculate the second conditional probability, which is the probability of the future value of the wind speed error given the historical wind speed error vector and the environmental feature vector: In the formula, This represents the joint probability of the environmental feature vector, the historical wind speed error vector, and the future wind speed error value. This represents the joint probability of the environmental feature vector and the historical wind speed error vector.
[0070] Based on the joint probability, the first conditional probability, and the second conditional probability mentioned above, calculate the transfer entropy of environmental characteristic data to wind speed error data: In the formula, This represents the environmental characteristic of cabinet door opening data; This indicates wind speed error data; This represents the propagation entropy of the cabinet door opening data to the wind speed error data. The larger the propagation entropy, the more additional information the historical information of the cabinet door opening data contributes to the future changes in the wind speed error data, meaning the higher the importance of the cabinet door opening data to the wind speed error data.
[0071] The transfer entropy of cabinet door opening data to wind speed error data is used as the dynamic correlation value between cabinet door opening data and wind speed error data, as specifically expressed below: Similarly, the dynamic correlation between air density data and wind speed error data, and the dynamic correlation between the air pressure difference data inside and outside the cabinet and wind speed error data can be calculated.
[0072] Calculate the comprehensive correlation value between each type of environmental characteristic data and wind speed error data.
[0073] Since the numerical ranges of static and dynamic correlation values may differ (e.g., mutual information may be 0-2, and propagation entropy may be 0-5), they need to be standardized to the [0,1] interval first.
[0074] Taking the static correlation value of cabinet door opening data as an example, the static correlation value is standardized as follows: In the formula, This represents the minimum static correlation value corresponding to three environmental feature data. This represents the maximum value of the static correlation value corresponding to the three environmental feature data.
[0075] Taking the dynamic correlation value of cabinet door opening data as an example, the dynamic correlation value is standardized as follows: In the formula, This represents the minimum value of the dynamic correlation value corresponding to the three environmental feature data. This represents the maximum value of the dynamic correlation value corresponding to the three environmental feature data.
[0076] The formula for calculating the overall correlation value is as follows: in, The weight of the static correlation value is set to 0.5 in this embodiment, but it can be adjusted according to specific circumstances.
[0077] Similarly, the combined correlation value between air density data and wind speed error data can be calculated. The comprehensive correlation value between the air pressure difference data inside and outside the cabinet and the wind speed error data. .
[0078] S3. Train a random forest model based on historical sample data and corresponding comprehensive correlation values. During the model training process, the weight of each decision tree is calculated.
[0079] Determine forest size: Set the total number of decision trees ( =50-200), each tree randomly selects one to three of three environmental feature data to form a fixed combination. When all nodes of the whole tree are split, only the corresponding environmental feature data is selected from this combination.
[0080] Training data: Training samples are extracted from all historical sample data, including three types of environmental feature data and corresponding labels (wind speed error data), with a sample size of no less than 1000 groups.
[0081] Decision tree construction and training: For each tree, bootstrap sampling is used to extract samples (with replacement) from the training set. The decision tree is constructed based on the CART algorithm (classification and regression tree). The optimal splitting node is selected by variance reduction, and finally, the error prediction value of a single tree is output. ( The tree number, 1≤ ≤ , (Total number of decision trees).
[0082] For each decision tree The matching degree is calculated based on the combination of the types of environmental feature data input. If decision tree The types of input environmental feature data are: (In this embodiment) ),but: In the formula, This represents the number of categories of environmental characteristic data; Indicates the first Environmental characteristic data; Indicates the first The comprehensive correlation value corresponding to various environmental characteristic data.
[0083] The formula for calculating the weight of each decision tree is as follows: In the formula, Indicates the first The matching degree of each decision tree; the matching degree is normalized to the tree weights. Ensure that the sum of the weights is 1.
[0084] Each decision tree randomly selects types of environmental feature data and historical sample data, increasing the model's randomness and reducing the risk of overfitting. Furthermore, different trees use different combinations of environmental feature data types, increasing model diversity and improving the overall model's generalization ability. The matching degree of each decision tree is calculated. It can evaluate the contribution of different combinations of environmental feature data to the model, using normalized matching degrees as weights. This ensures that the contribution of each decision tree is proportional to its matching degree, thus improving the model's interpretability.
[0085] S4. Input the real-time environmental feature data into the trained model, output the predicted wind speed error of a single tree, and then sum the weighted values to obtain the final predicted wind speed error value. Adjust the opening of the real-time air volume control valve according to the final predicted wind speed error value.
[0086] The formula for calculating the final wind speed error prediction is as follows: In the formula, Indicates the tree number; Indicates the total number of decision trees; Indicates the first The weights of each decision tree; Indicates the first The wind speed prediction error of each decision tree.
[0087] The process of adjusting the real-time airflow control valve opening based on the final predicted wind speed error is as follows: set a fixed proportional coefficient, calculate the product of the negative of the final predicted wind speed error and the fixed proportional coefficient, calculate the sum of this product and the real-time airflow control valve opening to obtain the target opening, and adjust the real-time airflow control valve opening according to the target opening.
[0088] The formula for calculating the target opening is as follows: In the formula, Indicates the target opening degree of the air volume control valve; This indicates the target opening degree of the airflow control valve at the current moment; , This represents a fixed proportional coefficient, preset according to the characteristics of the fume hood system, for example, 10, with units of % / (m / s). That is, every 1 m / s error corresponds to a 10% adjustment of the opening. The negative sign indicates that the adjustment direction is opposite to the error. When the value is positive (the measured surface wind speed is higher than the target wind speed), If it is negative, reduce the control valve opening to reduce the face velocity. When it is negative (the measured surface wind speed is lower than the target wind speed), A positive value increases the control valve opening to reduce face velocity; the clip function is used to limit the opening to an effective range of 0% to 100%.
[0089] By inputting real-time environmental feature data into a pre-trained random forest model, the predicted wind speed error of each tree is output, and then weighted and summed to obtain the final predicted wind speed error value. This effectively improves prediction accuracy and system reliability. Adjusting the opening of the real-time airflow control valve based on the final predicted wind speed error value ensures that the face velocity inside the fume hood remains stable within a safe range, demonstrating significant practical value.
[0090] 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 controlling the airflow of a fume hood using a control valve, characterized in that, include: The system acquires the opening degree of the fume hood's airflow control valve in real time and adjusts the valve opening based on the fume hood's environmental characteristic data to stabilize the face velocity inside the fume hood within a safe range. The environmental characteristic data includes door opening data, air density data, and the pressure difference between the inside and outside of the fume hood. The system monitors the fume hood door opening data in real time using a displacement sensor and acquires air density data. In the formula, This represents air density; P represents total air pressure, obtained from a pressure sensor. This indicates the partial pressure of water vapor, and the unit is Pa (Pascal). , Relative humidity is indicated by a temperature and humidity sensor. This represents the saturated water vapor pressure, obtained by looking up the thermodynamic temperature in a table. Represents the water vapor gas constant; The adjustment process includes: collecting data from a preset time period prior to the current moment and preprocessing it to obtain historical sample data. The historical sample data includes environmental characteristic data and wind speed error data. The wind speed error data is the difference between the measured surface wind speed and the target surface wind speed at the time of collection. The comprehensive correlation value between each type of environmental characteristic data and the wind speed error data is calculated. The calculation process for the comprehensive correlation value is as follows: calculating the static correlation value and dynamic correlation value between each type of environmental characteristic data and the wind speed error data, and then weighted summing the corresponding static and dynamic correlation values to obtain the comprehensive correlation value between each type of environmental characteristic data and the wind speed error data. The calculation process for the static correlation value is as follows: For each type of environmental feature data, the environmental feature data and wind speed error data in all historical sample data are discretized, and discrete signs are assigned to the discretized intervals. The joint probability distribution and marginal probability distribution of the discrete signs of the environmental feature data and the wind speed error data are calculated. Based on the joint probability distribution and marginal probability distribution, the mutual information value of the environmental feature data and the wind speed error data is calculated to obtain the static correlation value. Alternatively, for each type of environmental feature data, the environmental feature data in all historical sample data is binned, and the wind speed error data in the historical sample data of the environmental feature data in each bin are statistically analyzed. The variance of the wind speed error within each bin and the range of the wind speed error between all bins are calculated. The ratio of the weighted sum of the variances of the wind speed error within each bin to the range of the wind speed error between each bin is used as the static correlation value. A random forest model is trained based on the historical sample data and the corresponding comprehensive correlation value. During the model training process, the weight of each decision tree is calculated. The calculation process of the weight of each decision tree is as follows: For each decision tree, the comprehensive correlation value between all types of environmental feature data and wind speed error data is obtained, and the average of all comprehensive correlation values is taken as the matching degree of the decision tree. The ratio of this matching degree to the sum of the matching degrees of all decision trees is taken as the weight of the decision tree. Real-time environmental feature data is input into the trained model, and the predicted wind speed error of a single tree is output. Then, the weighted sum is obtained to obtain the final predicted wind speed error value. The opening of the real-time air volume control valve is adjusted according to the final predicted wind speed error value.
2. The airflow control valve control method based on a fume hood according to claim 1, characterized in that, The calculation process for the dynamic correlation value is as follows: All historical sample data are arranged in chronological order of collection to form a historical sample sequence. For each type of environmental characteristic data, the absolute value of the Pearson correlation coefficient between the environmental characteristic data and the wind speed error data within the collection period is used as the dynamic correlation value.
3. The airflow control valve control method based on a fume hood according to claim 1, characterized in that, The calculation process for the dynamic correlation value is as follows: All historical sample data are arranged in the order of collection time to form a historical sample sequence. For each type of environmental feature data, state sample data is constructed based on the collection time. The state sample data includes environmental feature vector, historical wind speed error vector, and future wind speed error value. The transfer entropy of environmental feature data to wind speed error data is calculated based on the state sample data to obtain the dynamic correlation value.
4. The airflow control valve control method based on a fume hood according to claim 3, characterized in that, The calculation process for the transfer entropy is as follows: The equal-frequency binning method is used to divide the environmental feature vector, historical wind speed error vector, and future wind speed error value of all state sample data into multiple intervals, and a discrete symbol is assigned to each interval. Based on the discrete symbol, the joint probability, the first conditional probability, and the second conditional probability of the environmental feature vector, the historical wind speed error vector, and the future wind speed error value are calculated. The transfer entropy of the environmental feature data to the wind speed error data is calculated based on the joint probability, the first conditional probability, and the second conditional probability.
5. The airflow control valve control method based on a fume hood according to claim 1, characterized in that, The process of adjusting the real-time airflow control valve opening based on the final wind speed error prediction value is as follows: set a fixed proportional coefficient, calculate the product of the negative of the final wind speed error prediction value and the fixed proportional coefficient, calculate the sum of the product and the real-time airflow control valve opening to obtain the target opening, and adjust the real-time airflow control valve opening according to the target opening.