Suspension chain cleaning control method based on interval optimization

Abstract

The invention relates to a suspension chain cleaning control method based on interval optimization, which comprises the following steps of: firstly, constructing an enhanced input vector containing organic, inorganic and particle stain degrees and composite characteristics thereof, training a plurality of support vector regression models based on historical data, and establishing nonlinear mapping between the stain degrees and cleaning process parameters; a dual-path dynamic weight calculation mechanism is innovatively introduced, and a linear extraction or arrangement importance method is adaptively selected according to data linear correlation strength to determine a feature weight; in the online control stage, a self-adaptive process parameter interval is dynamically generated by calculating the distance between a real-time fouling feature and a model support vector, and a final execution parameter is intelligently optimized in the interval based on a feature weight, so that precise self-adaptive cleaning aiming at different fouling degrees is realized; according to the scheme, the problems of resource waste and insufficient cleaning in traditional fixed program cleaning are effectively solved, and the intelligent level and resource utilization efficiency of the cleaning process are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of cleaning machine technology, specifically to a suspended chain cleaning control method based on interval optimization; Background Technology

[0002] Overhead chain systems are widely used in surface treatment workshops in industries such as automobiles and home appliances, where cleanliness directly affects product quality. Traditional cleaning control methods often employ switch control based on fixed thresholds or simple PID control. While these methods are logically simple, they suffer from slow response and cannot adapt to dynamic changes in working conditions, easily leading to insufficient or excessive cleaning. To further improve energy efficiency, the existing technology has introduced a range optimization control method. This method sets a target range for cleaning intensity and seeks the best balance between energy consumption and cleaning effect within the range. However, the optimization range of this type of method is usually static and lacks adaptability. When the system is under special operating conditions, the static range cannot make forward-looking adjustments, resulting in a decrease in control performance. In recent years, machine learning algorithms such as support vector machines (SVM) have been attempted for state identification in industrial processes; however, existing applications mostly treat machine learning models as independent monitoring modules, and their identification results are only connected to the control system through simple rules. This loose coupling fails to fully leverage the deep decision-making value of machine learning models and cannot achieve online adaptive adjustment of the core parameters of the control system. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a suspended chain cleaning control method based on interval optimization. This method has the advantage of enabling online adaptive adjustment of cleaning parameters for the core parameters of the control system. It solves the problem that traditional cleaning control methods often use on-off control based on fixed thresholds or simple PID control. While these methods are logically simple, they have a slow response and cannot adapt to dynamic changes in working conditions, which can easily lead to insufficient or excessive cleaning. To achieve the above objectives, the present invention provides the following technical solution: a suspended chain cleaning control method based on interval optimization, comprising the following steps: S1: Detect surface contamination on the workpiece and construct a multivariate feature vector of the workpiece's contamination level based on the detection data. and on the feature vector Feature enhancement is performed to obtain feature vectors. ; S2: Based on historical experience, the optimal process parameters for different levels of contamination are verified to be optimal, and a target vector of multiple process parameters for the cleaning process is constructed. ; S3: For the collected feature vectors With the target vector Normalization preprocessing is performed to obtain a standardized dataset for model training. ; S4: Using standardized datasets eigenvectors in Input and target vectors For the output, train a set of support vector regression models; S5. Construct a dual-path dynamic weighting mechanism based on the trained support vector regression model and set a judgment threshold. Based on the judgment result, different paths are selected to obtain the weight vector. ; S6: Real-time acquisition of the degree of contamination of the workpiece to be cleaned. The degree of turbidity was obtained after pretreatment. The values ​​are input into the trained support vector regression model to obtain the initial predicted values ​​of various process parameters. ; S7: Based on the initial predicted value By combining a preset range optimization strategy, an optimal operating range that can be flexibly selected is generated for each process parameter. ; S8: The cleaning machine controller receives the optimal operating range. According to the weight vector and turbidity parameters A judgment strategy is constructed, the final execution parameters are selected based on the judgment results, and the cleaning machine is driven to perform adaptive cleaning operations; Furthermore, in step S1, the multivariate feature vector ,in These represent the normalized levels of organic pollution, inorganic pollution, and particulate pollution, respectively. The target vector of multiple process parameters for: The parameters correspond to the pre-cleaning spray pressure, pre-cleaning time, ultrasonic cleaning power, ultrasonic cleaning time, rinsing flow rate, rinsing time, drying air volume, and drying time, respectively. The multivariate feature vector in step S1 Perform feature enhancement: in, , used to represent the interaction between characteristic organic and inorganic stains; , used to represent the saturation effect of particulate stains; Used to indicate the overall pollution level; In step S3, the training dataset D is composed of... The sample consists of: Each of them and The corresponding optimal combination of process parameters, verified by experiments.

[0004] Furthermore, step S4 includes the following sub-steps: S4.1. Employ a multi-output support vector regression strategy, with the target vector as the target vector. For each process parameter, an independent support vector regression model is established. ; S4.2 Each model uses radial basis functions as kernel functions and solves the optimization problem of insensitive support vector regression. S4.3, From Augmented Dataset Constructing a sub-training set Define and solve the optimization problem for the k-th SVR model. For the k-th process parameter, the final decision function is obtained: in For Lagrange multipliers, Let b be the kernel function and b be the bias term.

[0005] Furthermore, step S5 also includes the following sub-steps: S5.1 Calculate the feature-parameter correlation matrix R, where the element Rjk is the Pearson correlation coefficient between feature j and parameter k; S5.2 Calculate the average absolute correlation coefficient : S5.3 Constructing the judgment threshold And based on the judgment threshold The judgment conditions are as follows: like This indicates a strong linear correlation, so path A should be chosen. like This indicates that nonlinearity is dominant, and path B should be chosen; S5.4. Based on the judgment results, assign the final dynamic weight vector to each process parameter k. ,in .

[0006] Furthermore, path A in S5.4 includes the following calculation steps: A.1. For each parameter k, use Train a linear kernel SVR model; A.2. Directly extract the weight coefficient vector from the trained linear SVR model. ; A.3, the weight vector This refers to the dynamic weights of each feature on each process parameter k under path A. .

[0007] The path B in S5.5 includes the following calculation steps: B.1 Using the SVR model already trained in S4.3 Using the corresponding validation dataset, calculate a baseline performance score for the SVR model on the validation set. ; B.2. For the j-th feature, randomly shuffle the validation set for that feature. All values, destroying their relationship with the true value Relationship; B.3. Using the shuffled validation set, apply the model. Make predictions and calculate new performance scores. ; B.4 Calculate the importance score of this feature. in Compared to The greater the decrease, the more important the feature; B.5. Repeat steps B.2-B.4 for all features to obtain the importance feature vector: B.6, Yes Normalization is performed so that the sum of all components is 1, resulting in the dynamic weight vector for path B. .

[0008] Furthermore, step S6 includes the following sub-steps: S6.1, Use the same method as in step S1 to adjust the turbidity parameter. Feature enhancement is performed to obtain the degree of turbidity parameters. ; S6.2, Regarding the turbidity parameter Normalization was performed to obtain the turbidity parameter. ; S6.3, The turbidity level parameter Output 8 SVR models Obtain normalized predicted values ; S6.4, Regarding the predicted value The predicted values ​​of physical quantities are obtained by performing inverse normalization. .

[0009] Furthermore, generating the optimal operating range in step S7 includes the following steps: S7.1 Calculate each current sample The minimum feature space distance dk to the support vector set of model fk; S7.2 Calculate the adaptive tolerance for each parameter: in Basic tolerance, This is the scaling factor; S7.3, Generate adaptive optimal operating range: .

[0010] Furthermore, in step S8, the final execution parameters are selected within the optimal operation range, and the specific determination strategy is as follows: S8.1, Controller receives adaptive parameter range ; S8.2. Based on the dynamic weight vector Wk, for each process parameter k and its interval... Calculate the weighted sum of features, Sk: S8.3, Based on the weighted sum Sk and the threshold Compare the results to determine the final execution value: in The upper limit of the interval, The lower limit of the interval, It is the midpoint of the interval.

[0011] Compared with the prior art, the technical solution of this application has the following beneficial effects: I. This suspended chain cleaning control method based on interval optimization solves the problems of "insufficient cleaning" or "over-cleaning" under traditional fixed programs. The system can accurately identify different combinations and degrees of organic, inorganic, and particulate stains, and match the most suitable cleaning intensity and time. For example, for workpieces with high particle content and low organic content, it will automatically increase the spray pressure of pre-cleaning while appropriately reducing the ultrasonic power, realizing "tailored measures according to contamination". This increases the cleaning qualification rate (such as the proportion of workpieces that meet a specific cleanliness standard) to nearly 100% and greatly reduces rework caused by cleaning problems. Second, this suspended chain cleaning control method based on interval optimization realizes the transformation from "extensive" cleaning to "refined" cleaning. For lightly contaminated workpieces, the system automatically adjusts parameters such as spray pressure, ultrasonic power, and drying air volume, and shortens the process time. Actual measurement data shows that, while ensuring cleaning quality, the system can save an average of 15% to 30% of water consumption, 10% to 25% of electrical energy (mainly from the ultrasonic and drying processes), and corresponding cleaning agent consumption, directly reducing production costs. Third, the overhead chain cleaning control method based on interval optimization no longer "malfunctions" when faced with unprecedented, mixed-type, or extremely dirty workpieces; the adaptive interval optimization mechanism can perceive the difference between the current sample and historical experience, and maintain the robustness of operation by expanding the parameter range, preventing the scrapping of the entire batch of workpieces due to a single erroneous prediction; at the same time, the dual-path weight calculation ensures that the most suitable decision basis can be found under any data characteristics, enabling the system to show strong adaptability when facing complex and ever-changing real production environments; Fourth, this suspended chain cleaning control method based on interval optimization changes the traditional control "black box" or "experience-based" mode; the system provides not a single value, but an optimal parameter range, and provides clear data basis based on feature weights for the final selection within the range; this provides process engineers with a flexible decision space to optimize production strategies (such as prioritizing energy saving or efficiency), and at the same time makes the logic of the entire control process clear, explainable and traceable. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the process of the present invention; Figure 2 This is a schematic diagram illustrating the path determination method of the present invention; Figure 3 This is a flowchart illustrating the execution process of path A in this invention. Figure 4 This is a diagram illustrating the execution process of path B in this invention. Figure 5 This is a diagram illustrating the execution process within the optimal parameter range of this invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Please see the appendix Figure 1-5This invention provides a suspended chain cleaning control method based on interval optimization, comprising the following steps: S1: Detect surface contamination on the workpiece and construct a multivariate feature vector of the workpiece's contamination level based on the detection data. and on the feature vector Feature enhancement is performed to obtain feature vectors. ; S2: Based on historical experience, the optimal process parameters for different levels of contamination are verified to be optimal, and a target vector of multiple process parameters for the cleaning process is constructed. ; S3: For the collected feature vectors With the target vector Normalization preprocessing is performed to obtain a standardized dataset for model training. ; S4: Using standardized datasets eigenvectors in Input and target vectors For the output, train a set of support vector regression models; S5. Construct a dual-path dynamic weighting mechanism based on the trained vector regression model and set judgment conditions. Select different paths to obtain the weight vector based on the judgment results. ; S5: Real-time acquisition of the degree of contamination of the workpiece to be cleaned. The degree of turbidity was obtained after pretreatment. The values ​​are input into the trained support vector regression model to obtain the initial predicted values ​​of various process parameters. ; S6: Based on the initial predicted value By combining a preset range optimization strategy, an optimal operating range is generated for each process parameter, which can be flexibly selected. ; S7: The cleaning machine controller receives the optimal operating range. According to the weight vector and turbidity parameters A judgment strategy is constructed, the final execution parameters are selected based on the judgment results, and the cleaning machine is driven to perform adaptive cleaning operations; In step S1, the multivariate feature vector ,in These represent the normalized levels of organic pollution, inorganic pollution, and particulate pollution, respectively. The target vector of multiple process parameters for: ; Its parameters correspond to the pre-cleaning spray pressure, pre-cleaning time, ultrasonic cleaning power, ultrasonic cleaning time, rinsing flow rate, rinsing time, drying air volume, and drying time, respectively; it clarifies the objectives that the model needs to learn and predict; it covers the key controllable parameters of the entire process from pre-cleaning to drying, ensuring that the optimization control is global and systematic, rather than targeting a single link, thereby guaranteeing the final overall cleaning effect; The multivariate feature vector in step S1 Perform feature enhancement: , in, , used to represent the interaction between characteristic organic and inorganic stains; , used to represent the saturation effect of particulate stains; Used to indicate the overall pollution level; This step is the cornerstone of data modeling. First, three basic stain types are defined as core inputs. The subsequent feature engineering is crucial. By introducing interaction terms, nonlinear terms, and comprehensive indices between stains, it transforms single, isolated stain information into a composite feature vector that can characterize the synergistic effect, saturation effect, and overall load between stains. This greatly enriches the information dimension that the model can learn and lays the foundation for accurately modeling complex cleaning dynamics relationships in the future. In step S3, the training dataset D is composed of... The sample consists of: , Each of them and The corresponding optimal combination of process parameters, verified by experiments; High-quality datasets are essential for model performance; this step systematically collects historical best-performing data as “teacher signals”; the key optimization lies in using enhanced feature vectors to construct the dataset, which means that the model will learn from the beginning a deeper mapping relationship based on composite features, rather than a shallow linear association. S4 includes the following sub-steps: S4.1. Employ a multi-output support vector regression strategy, with the target vector as the target vector. For each process parameter, an independent support vector regression model is established. ; S4.2 Each model uses radial basis functions as kernel functions and solves the optimization problem of insensitive support vector regression. S4.3, From Augmented Dataset Constructing a sub-training set Define and solve the optimization problem for the k-th SVR model. For the k-th process parameter, the final decision function is obtained: in For Lagrange multipliers, Here, b is the kernel function and b is the bias term; this is the core training process of the model; each SVR model learns how to predict the optimal single process parameter from the complex fouling features by solving a constrained optimization problem; the final decision function is the mathematical expression of the "fouling-parameter" mapping relationship. S4 further includes the following sub-steps: S4.4 Construct judgment conditions, and select different paths based on the judgment results to calculate the final dynamic weight vector Wk for each parameter k; S4.5 Calculate the feature-parameter correlation matrix R, where the element Rjk is the Pearson correlation coefficient between feature j and parameter k; S4.6 Calculate the average absolute correlation coefficient : S4.7 Constructing the judgment threshold And based on the judgment threshold The judgment conditions are as follows: like This indicates a strong linear correlation, so path A should be chosen. like This indicates that nonlinearity is dominant, and path B should be chosen; Among them, threshold To determine the overall linearity of a dataset, its determination can be based on statistical correlation significance tests. A scientific approach is to use the framework of permutation tests: first, randomly shuffle the correspondence between features and labels in the original dataset multiple times (e.g., 1000 times), and calculate the mean absolute correlation coefficient after each shuffle. This leads to a hypothesis under the null hypothesis (features are independent of labels). Experience distribution. Then, the true distribution is calculated from the original data. Compared to this distribution, the threshold It can be set to a high quantile of this empirical distribution (such as the 95th or 99th quantile). If If the observed linear correlation exceeds this threshold, it indicates that the observed linear correlation is statistically significant, thus selecting path A. This method ensures that path selection is not based on an arbitrary fixed value, but rather on statistical significance.

[0014] S4.8. Based on the judgment results, assign the final dynamic weight vector to each process parameter k. ,in, ; The path A in S4.7 includes the following calculation steps: A.1. For each parameter k, use Train a linear kernel SVR model; A.2. Directly extract the weight coefficient vector from the trained linear SVR model. ; A.3, the weight vector This refers to the dynamic weights of each feature on each process parameter k under path A. The path B in S4.7 includes the following calculation steps: B.1 Using the SVR model already trained in S4.3 Using the corresponding validation dataset, calculate a baseline performance score for the SVR model on the validation set. ; B.2. For the j-th feature, randomly shuffle the validation set for that feature. All values, destroying their relationship with the true value Relationship; B.3. Using the shuffled validation set, apply the model. Make predictions and calculate new performance scores. ; B.4 Calculate the importance score of this feature. in Compared to The greater the decrease, the more important the feature; B.5. Repeat steps B.2-B.4 for all features to obtain the importance feature vector: B.6, Yes Normalization is performed so that the sum of all components is 1, resulting in the dynamic weight vector for path B. ; This step is a key enhancement to the intelligence of this solution; it is not a fixed calculation, but an adaptive decision-making process based on self-assessment of data characteristics. By analyzing the overall linear correlation between features and parameters, the system automatically selects the most appropriate weight interpretation method: for datasets with strong linear relationships, efficient and intuitive linear weights are used; for datasets dominated by nonlinearity, a more robust and more reflective ranking of importance is used. This ensures that the weights relied upon for subsequent control decisions are the most accurate and reliable. S5 includes the following sub-steps: S5.1, Use the same method as in step S1 to adjust the turbidity parameter. Feature enhancement is performed to obtain the degree of turbidity parameters. ; S5.2, Regarding the turbidity parameter Normalization was performed to obtain the turbidity parameter. ; During online operation, the feature processing flow from offline training is fully replicated for new workpieces. This not only ensures the consistency of model input but, more importantly, transforms raw, coarse dirtiness readings into information-rich composite feature vectors in real time, providing high-quality input for accurate prediction. S5.3, The turbidity level parameter Output 8 SVR models Obtain normalized predicted values ; By utilizing a pre-trained ensemble of models, parallel reasoning is performed on the complex contamination characteristics of the current workpiece to quickly obtain a set of preliminary, personalized process parameter recommendations; this is the first step from "perception" to "decision-making". S5.4, For predicted values The predicted values ​​of physical quantities are obtained by performing inverse normalization. ; The normalized values ​​output by the model are restored to parameter values ​​with actual physical meaning so that the control system can recognize and execute them. The process of generating the optimal operating range in step S6 includes the following steps: S6.1 Calculate each current sample The minimum feature space distance dk to the support vector set of model fk; S6.2 Calculate the adaptive tolerance for each parameter: in Basic tolerance, This is the scaling factor; S6.3, Generate adaptive optimal operation range: This step transforms a single, rigid prediction into a flexible, robust operating range. Its innovation lies in the introduction of an "uncertainty metric": when a new workpiece is similar to historical experience, the range narrows, and precise control is executed; when encountering an unfamiliar contamination pattern, the range widens, avoiding cleaning failures caused by the model's "blind confidence," and greatly enhancing the system's adaptability and reliability under complex working conditions.

[0015] In step S7, the final execution parameters are selected within the optimal operation range. The specific determination strategy is as follows: S7.1, Controller receives adaptive parameter range ; S7.2. Based on the dynamic weight vector Wk, for each process parameter k and its interval... Calculate the weighted sum of features, Sk: ; S7.3, Based on the weighted sum Sk and the threshold Compare the results to determine the final execution value: in The upper limit of the interval, The lower limit of the interval, It is the midpoint of the interval; Where the threshold This is used for intelligent optimization within a parameter range, and its determination can be based on a weighted sum of weights and features. The historical distribution of . By analyzing the training set or a reserved validation set, for each process parameter Calculate all samples Values ​​and analyze their distribution. Threshold These can be set to specific quantiles at the positive and negative ends of the distribution, for example, setting the positive threshold... Set as 75th percentile, negative threshold Set it to the 25th percentile. This means that only when the current sample's... An aggressive strategy using the upper and lower limits of a range is triggered only when the value falls within the highest or lowest 25% range based on historical experience; otherwise, a conservative median value is used. This method links the decision threshold to the inherent distribution of historical data, ensuring that the decision-making behavior aligns with known data patterns and is more robust.

[0016] This is the final stage of control decision-making, which solves the problem of "given an optimal range, which value is the best to choose"; by using dynamic weights, the current pollution characteristics are weighted and evaluated, so as to make the most targeted fine-tuning within the optimal range; for example, even if the overall pollution is not heavy, if the particulate matter characteristics and their weights are abnormally prominent, the system will also select the upper limit value within the spray pressure range to achieve true intelligent and refined control.

[0017] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A suspended chain cleaning control method based on interval optimization, characterized in that, Includes the following steps: S1: Detect surface contamination on the workpiece and construct a multivariate feature vector of the workpiece's contamination level based on the detection data. and on the feature vector Feature enhancement is performed to obtain feature vectors. ; S2: Based on historical experience, the optimal process parameters for different levels of contamination are verified to be optimal, and a target vector of multiple process parameters for the cleaning process is constructed. ; S3: For the collected feature vectors With the target vector Normalization preprocessing is performed to obtain a standardized dataset for model training. ; S4: Using standardized datasets eigenvectors in Input and target vectors For the output, train a set of support vector regression models; S5. Construct a dual-path dynamic weighting mechanism based on the trained support vector regression model and set a judgment threshold. Based on the judgment result, different paths are selected to obtain the weight vector. ; S6: Real-time acquisition of the degree of contamination of the workpiece to be cleaned. The degree of turbidity was obtained after pretreatment. The values ​​are input into the trained support vector regression model to obtain the initial predicted values ​​of various process parameters. ; S7: Based on the initial predicted value By combining a preset range optimization strategy, an optimal operating range that can be flexibly selected is generated for each process parameter. ; S8: The cleaning machine controller receives the optimal operating range. According to the weight vector and turbidity parameters A judgment strategy is constructed, the final execution parameters are selected based on the judgment results, and the cleaning machine is driven to perform adaptive cleaning operations.

2. The suspended chain cleaning control method based on interval optimization according to claim 1, characterized in that: In step S1, the multivariate feature vector ,in These represent the normalized levels of organic pollution, inorganic pollution, and particulate pollution, respectively. The target vector of multiple process parameters for: ； The parameters correspond to the pre-cleaning spray pressure, pre-cleaning time, ultrasonic cleaning power, ultrasonic cleaning time, rinsing flow rate, rinsing time, drying air volume, and drying time, respectively. The multivariate feature vector in step S1 Perform feature enhancement: ， in, , used to represent the interaction between characteristic organic and inorganic stains; , used to represent the saturation effect of particulate stains; Used to indicate the overall pollution level; In step S3, the training dataset D is composed of... The sample consists of: ， Each of them and The corresponding optimal combination of process parameters, verified by experiments.

3. The suspended chain cleaning control method based on interval optimization according to claim 1, characterized in that, S4 includes the following sub-steps: S4.

1. Employ a multi-output support vector regression strategy, with the target vector as the target vector. For each process parameter, an independent support vector regression model is established. ; S4.2 Each model uses radial basis functions as kernel functions and solves the optimization problem of insensitive support vector regression. S4.3, From Augmented Dataset Constructing a sub-training set Define and solve the optimization problem for the k-th SVR model. For the k-th process parameter, the final decision function is obtained: in For Lagrange multipliers, Let b be the kernel function and b be the bias term.

4. The suspended chain cleaning control method based on interval optimization according to claim 1, characterized in that, S5 further includes the following sub-steps: S5.1 Calculate the feature-parameter correlation matrix R, where the element Rjk is the Pearson correlation coefficient between feature j and parameter k; S5.2 Calculate the average absolute correlation coefficient : S5.3 Constructing the judgment threshold And based on the judgment threshold The judgment conditions are as follows: like This indicates a strong linear correlation, so path A should be chosen. like This indicates that nonlinearity is dominant, and path B should be chosen; S5.

4. Based on the judgment results, assign the final dynamic weight vector to each process parameter k. ,in 。 5. The suspended chain cleaning control method based on interval optimization according to claim 4, characterized in that, The path A in S5.4 includes the following calculation steps: A.

1. For each parameter k, use Train a linear kernel SVR model; A.

2. Directly extract the weight coefficient vector from the trained linear SVR model. ; A.3, the weight vector This refers to the dynamic weights of each feature on each process parameter k under path A. The path B in S5.5 includes the following calculation steps: B.1 Using the SVR model already trained in S4.3 Using the corresponding validation dataset, calculate a baseline performance score for the SVR model on the validation set. ; B.

2. For the j-th feature, randomly shuffle the validation set for that feature. All values, destroying their relationship with the true value Relationship; B.

3. Using the shuffled validation set, apply the model. Make predictions and calculate new performance scores. ; B.4 Calculate the importance score of this feature. in Compared to The greater the decrease, the more important the feature; B.

5. Repeat steps B.2-B.4 for all features to obtain the importance feature vector: B.6, Yes Normalization is performed so that the sum of all components is 1, resulting in the dynamic weight vector for path B. .

6. The suspended chain cleaning control method based on interval optimization according to claim 1, characterized in that: S6 includes the following sub-steps: S6.1, Use the same method as in step S1 to adjust the turbidity parameter. Feature enhancement is performed to obtain the degree of turbidity parameters. ; S6.2, Regarding the turbidity parameter Normalization was performed to obtain the turbidity parameter. ; S6.3, The turbidity level parameter Output 8 SVR models Obtain normalized predicted values ; S6.4, Regarding the predicted value The predicted values ​​of physical quantities are obtained by performing inverse normalization. .

7. The suspended chain cleaning control method based on interval optimization according to claim 1, characterized in that: The process of generating the optimal operating range in S7 includes the following steps: S7.1 Calculate each current sample The minimum feature space distance dk to the support vector set of model fk; S7.2 Calculate the adaptive tolerance for each parameter: in Basic tolerance, This is the scaling factor; S7.3, Generate adaptive optimal operating range: 。 8. The suspended chain cleaning control method based on interval optimization according to claim 1, characterized in that: In step S8, the final execution parameter is selected within the optimal operation range. The specific determination strategy is as follows: S8.1, Controller receives adaptive parameter range ; S8.

2. Based on the dynamic weight vector Wk, for each process parameter k and its interval... Calculate the weighted sum of features, Sk: S8.3, Based on the weighted sum Sk and the threshold Compare the results to determine the final execution value: in The upper limit of the interval, The lower limit of the interval, It is the midpoint of the interval.

Images