Mooney viscosity soft measurement and process control method in mixing process of internal mixer
By establishing a Mooney online prediction model in the mixer and adjusting the mixing process using real-time data, the problem of Mooney viscosity control in the mixer is solved, and the quality and efficiency of the mixing glue are improved.
Patent Information
- Application Number
- CN202411144162.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-20
- Publication Date
- 2024-11-29
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
It is difficult to effectively improve the quality of the mixing glue during the mixing process of existing intensive mixers, especially due to the interference of external factors and the difficulty in controlling the Mooney's viscosity due to complex process.
Mooney's online prediction model is adopted to collect current, temperature, pressure, speed, energy and other parameters in real time, combine the Mooney's viscosity data of the quality monitoring system, establish and update the prediction model, and adjust the mixing process in real time to achieve the expected Mooney's viscosity .
The precise control of process parameters during the rubber mixing process is achieved, adverse effects and errors are reduced, and the quality and efficiency of the mixing rubber are improved.
Smart Images

Figure CN119036664A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of intelligent control of mixing of internal mixers, and in particular relates to a soft measurement and process control method for Mooney viscosity of a mixing process of an internal mixer. Background Art
[0002] The main indicators for judging the quality of internal mixer rubber are Mooney viscosity, carbon black dispersion, hardness, tensile strength, etc., with Mooney viscosity being the most important. The Mooney viscosity requirements vary from part to part of the tire, and it is impossible to test it in real time during the production process.
[0003] Due to the unstable fluidity of rubber, the different control time points at the end of mixing have a great influence on Mooney.
[0004] There are three main control methods commonly used for the control of the end point of the mixing cycle of internal mixers at home and abroad: time method, temperature method and energy method, or a combination of two variables. The general control method is mainly based on temperature, supplemented by the reference time method; or mainly based on energy, supplemented by temperature. The goal is to obtain uniform dispersion of the rubber in the shortest time, so that the physical properties and processing performance of the mixed rubber meet the expected requirements.
[0005] (1) Time method
[0006] The purpose of mixing is to make all components evenly dispersed, the physical and mechanical properties of the rubber are uniform and consistent, to ensure the performance of the final product; to improve the processing performance of the rubber and meet the processing requirements of the subsequent process. This requires a certain amount of time, neither insufficient time resulting in "under-mixing" nor too long time resulting in "over-mixing". Both "under-mixing" and "over-mixing" will affect the processing performance or physical and mechanical properties of the rubber. Therefore, the mixing time can reflect the quality of the rubber.
[0007] (2) Temperature method
[0008] In the internal mixer, the rubber compound comes into contact with the rotor, mixing chamber wall, discharge door and pressure weight surface. The friction, shearing and tearing processes will cause the rubber compound to heat up, and the temperature will continue to rise as the mixing process proceeds. The trend of the temperature curve basically rises from low to high. Therefore, the change in the quality of the rubber compound can be reflected by the temperature change.
[0009] (3) Energy method
[0010] The rubber mixing process is also the process of rubber absorbing energy. Under the condition of basically stable cooling conditions, the energy absorbed by unit volume of rubber during mixing can reflect the friction, shear and tearing process it obtains, so it can be measured by density.
[0011] The energy consumption of the mixing mill reflects the changes in rubber quality.
[0012] The three traditional control methods for the mixing process of the internal mixer mentioned above will be affected by external factors and affect the results, especially when the plasticity of the plasticized rubber fluctuates, even if the energy is used as the standard, it will affect the control effect. Due to the complexity of the mixing process and the diversity of the quality indicators of the mixed rubber, these methods have their limitations and will affect the Mooney viscosity and efficiency of the rubber. Summary of the invention
[0013] 1. Technical problem to be solved by the invention
[0014] The invention aims to solve the problem that it is difficult to effectively improve the quality of mixed rubber during the mixing process of the existing internal mixer.
[0015] 2. Technical solution
[0016] In order to achieve the above object, the technical solution provided by the present invention is:
[0017] The present invention provides a method for soft measurement and process control of Mooney viscosity in a mixing process of an internal mixer, the method comprising the following steps:
[0018] S100, establish Mooney online prediction model Model 门尼预测 ;
[0019] S200, real-time data X 过程实时 Input to the Mooney online prediction model Model established in step S100 门尼预测 Calculate the predicted Mooney viscosity value Y of the current mixing process 门尼预测 ;
[0020] S300, the predicted Mooney viscosity value Y obtained in step S200 门尼预测 and the preset Mooney expected value Y 门尼期望 Compare and calculate the absolute deviation. If the deviation is within the preset error range Err 范围 If the deviation is not within the preset error range Err, then step S400 is executed. 范围 If the number of steps is within 1, step S200 is repeated;
[0021] S400, send mixing process end sequence signal 结束 Control the internal mixer PLC to complete the mixing end operation.
[0022] Preferably, the step S100 specifically includes:
[0023] S110, collect data and create data set X 过程历史 and Y 门尼历史 ;
[0024] S120, the data set X obtained in step S100 过程历史and Y 门尼历史 Perform data cleaning and obtain a normal data set X 过程正常 and Y 门尼正常 ;
[0025] S130: the data set X obtained in step S120 过程正常 and Y 门尼正常 Genetic programming to generate new features X 过程特征合成 , and added to the original X 过程正常 To generate a new dataset X 过程候选 ;
[0026] S140, the data set X obtained in step S130 过程候选 The target Mooney viscosity Y is selected by the maximum correlation minimum redundancy algorithm. 门尼历史 The most relevant input feature X corr ;
[0027] S150: the input feature X obtained in step S140 corr and Y 门尼正常 Use extreme gradient boosting algorithm to achieve regression prediction and establish model Mooney prediction model 门尼预测 .
[0028] Preferably, the step S110 is specifically as follows:
[0029] Collect parameters during the mixing process to form a multivariate parameter data set X 过程历史 , define its corresponding independent variable X 样本 =(x1, x2, ..., x m ) T , where m is the number of parameters, and the entire data set X 过程历史 Then it is defined as X 过程历史 =(X 样本1 , X 样本2 , …, X 样本n ) where n is the number of cars in historical data; Mooney viscosity value corresponds to X 过程历史 , Y 门尼历史 =(y1, y2, ..., y n ) T , where n is the number of cars in the historical data; 门尼历史 Compared with the actual mixing process data X 过程历史 Align and complete the mapping of data labels and inputs.
[0030] Preferably, the step S120 is specifically as follows:
[0031] By using the Isolation Forest algorithm IsolationForest to remove X 过程历史 Neutralize Y 门尼历史 outliers in the dataset X to obtain the normal过程正常 and Y 门尼正常 ;
[0032] By constructing multiple Isolation Trees and calculating the average path length of each tree, the final anomaly score is obtained.
[0033] For a data point x, its path length in the tree is recorded as h(x);
[0034] The expected average path length can be approximated by the following formula:
[0035]
[0036] Where H(i) is the i-th harmonic number:
[0037]
[0038] Based on the comparison of path length with the average path length, an anomaly score can be defined:
[0039]
[0040] Among them, s(x, n) close to 1 indicates that the data point x is an outlier, and close to 0 indicates a normal point.
[0041] Preferably, the S130 is specifically to generate a new expression by combining existing features and operators, and select an expression with higher fitness as a new feature, wherein the fitness is the mixing process X 过程正常 Generate Pearson correlation coefficient between features and targets;
[0042] include
[0043] A. Initial population generation: define an operator set O and a terminator set T (variables and constants), randomly generate several expression trees, and the nodes of each tree are randomly selected from O and T;
[0044] B. Fitness evaluation, using the Pearson correlation coefficient to measure the linear correlation between the generated features and the target variable:
[0045]
[0046] Among them, x i is the generated eigenvalue, y i is the target variable value, and They are x i and i The fitness function can be defined as the absolute value of the correlation coefficient
[0047] f(x)=|r|
[0048] The higher the coefficient of correlation, the higher the fitness;
[0049] C. Selection: The selection operation determines which individuals will be used to generate the next generation based on their fitness. The probability of an individual being selected is proportional to its fitness, and the probability of selection pi can be expressed as:
[0050]
[0051] D. Crossover operation: The crossover operation exchanges part of the expression trees of two parent individuals to generate new offspring. The steps of the crossover operation are as follows:
[0052] Randomly select two parent individuals (A) and (B);
[0053] Randomly select a subtree from (A) and (B);
[0054] Interact these two subtrees to generate two new individuals;
[0055] E. Mutation operation: Mutation operation makes small-scale random modifications to individuals to increase the diversity of the population. The steps of mutation operation are as follows:
[0056] Randomly select an individual;
[0057] Randomly select a subtree from the individual;
[0058] Replace the subtree with a randomly generated subtree;
[0059] F. Termination: When any of the following termination conditions are met, the termination conditions include:
[0060] The preset maximum number of generations has been reached.
[0061] Find individuals whose fitness meets the characteristic number and exceeds the fitness threshold (0.9).
[0062] Fitness has not improved significantly over several generations;
[0063] G. Select features and select several expressions with the best performance as new features X according to fitness. 过程特征合成 , added to the original X 过程正常 To generate a new dataset X 过程候选 .
[0064] Preferably, the step S140 is specifically to obtain the input feature X through the following steps: corr ,include
[0065] a. Relevance, specifically
[0066] Use mutual information (MI) to define the relationship between the correlation measurement feature and the target variable, and define feature X i and the target variable Y, whose mutual information is defined as
[0067]
[0068] Among them, (p(x i ,y)) is (X i ) and (Y), (p(x i )) and (p(y)) are respectively (X i ) and (Y);
[0069] b. Redundancy, specifically
[0070] Mutual information is used to quantify the redundant information between redundancy measurement features. Given two features Xi and Xj, their redundant mutual information is defined as:
[0071]
[0072] Among them, p(x i ,x j ) is X i and X j The joint probability distribution of i ) and p(x j ) are X i and X j The marginal probability distribution of
[0073] c. mRMR objective function
[0074] Select a subset of features so that the correlation between these features and the target variable is maximized and the redundancy between these features is minimized. The objective function is defined as:
[0075]
[0076] Among them, S is the selected feature subset, |S| is the size of the subset;
[0077] d. Stepwise selection process
[0078] Initialization: Select the first feature so that it has the greatest correlation with the target variable Y;
[0079] Iteration selection:
[0080] At each step, a new feature Xk is selected so that the correlation between the feature and the target variable Y minus the redundancy between the feature and the currently selected feature subset is maximized, that is, the feature Xk selected at each step satisfies:
[0081]
[0082] Among them, Sk-1 is the feature subset that has been selected in the previous k-1 steps;
[0083] e. Termination conditions
[0084] The feature selection process usually terminates when the preset number of features or other stopping conditions are reached, and several variables are selected as the final modeling data set Xcorr.
[0085] Preferably, the S150 specifically includes
[0086] (1) Given a training data set X corr , Y 门尼正常 ;
[0087] (2) Initialize the model f0(x) to make it equal to the mean of all samples;
[0088] (3) For each iteration t = 1, 2, ..., T,
[0089] Calculate the negative gradient of the loss function as the residual of the current model, i.e., the pseudo residual
[0090] Use pseudo residuals to train a new decision tree and get the output value of the leaf node
[0091] Update the model Where η is the learning rate;
[0092] After completing the update iteration, the model Mooney prediction model Model is obtained 门尼预测 .
[0093] 3. Beneficial effects
[0094] Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects:
[0095] The present invention provides a method for soft measurement and process control of Mooney viscosity in a mixing process of an internal mixer, by collecting parameters X such as current, temperature, pressure, speed, energy, power, etc. during the rubber mixing process. 过程历史 , combined with the Mooney viscosity Y of the final rubber quality index collected by the quality monitoring system 门尼历史 , establish Mooney online prediction model Model 门尼预测 , and convert real-time data into 过程实时 Input to the Mooney online prediction model Model established in step S100 门尼预测 Calculate the predicted Mooney viscosity value Y of the current mixing process 门尼预测 ; Mooney viscosity predicted value Y门尼预测 By comparing with the preset Mooney expected value Y Mooney expectation and calculating its absolute deviation, it can effectively ensure that the process parameters in the actual rubber mixing process meet the expected settings, ensure the effect of rubber mixing, and reduce adverse effects and errors. BRIEF DESCRIPTION OF THE DRAWINGS
[0096] Figure 1 It is a schematic flow chart of a method for soft measurement and process control of Mooney viscosity in a mixing process of an internal mixer of the present invention;
[0097] Figure 2 Schematic diagram of offline training process of the model in this embodiment;
[0098] Figure 3 A schematic diagram of the flow of online prediction and control of the model of this embodiment. DETAILED DESCRIPTION
[0099] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.
[0100] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0101] In the present application, the terms "upper", "lower", "left", "right", "front", "back", "top", "bottom", "inner", "outer", "middle", "vertical", "horizontal", "lateral", "longitudinal" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings. These terms are mainly used to better describe the present application and its embodiments, and are not used to limit the indicated devices, elements or components to have a specific orientation, or to be constructed and operated in a specific orientation.
[0102] In addition, some of the above terms may be used to express other meanings in addition to indicating orientation or positional relationship. For example, the term "on" may also be used to express a certain dependency or connection relationship in some cases. For those of ordinary skill in the art, the specific meanings of these terms in this application can be understood according to specific circumstances.
[0103] In addition, the terms "installed", "set", "provided with", "connected", "connected", and "socketed" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection, or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be an internal connection between two devices, elements, or components. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.
[0104] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0105] Example 1
[0106] See attached Figure 1-3 , a method for soft measurement and process control of Mooney viscosity in a mixing process of an internal mixer according to this embodiment, the method comprises the following steps:
[0107] S100, establish Mooney online prediction model Model 门尼预测 ;
[0108] S200, real-time data X 过程实时 Input to the Mooney online prediction model Model established in step S100 门尼预测 Calculate the predicted Mooney viscosity value Y of the current mixing process 门尼预测 ;
[0109] S300, the predicted Mooney viscosity value Y obtained in step S200 门尼预测 and the preset Mooney expected value Y 门尼期望 Compare and calculate the absolute deviation. If the deviation is within the preset error range Err 范围 If the deviation is not within the preset error range Err, then step S400 is executed. 范围 If the number of steps is within 1, step S200 is repeated;
[0110] S400, send mixing process end sequence signal 结束 Control the internal mixer PLC to complete the mixing end operation.
[0111] The method of this embodiment uses the current, temperature, pressure, speed, energy, power and other parameters collected during the rubber mixing process to calculate the 过程历史 , combined with the Mooney viscosity Y in the final rubber quality index collected by the quality monitoring system 门尼历史 , establish Mooney online prediction model Model 门尼预测 , and convert real-time data into 过程实时 Input to the Mooney online prediction model Model established in step S100 门尼预测 Calculate the predicted Mooney viscosity value Y of the current mixing process 门尼预测 ; Mooney viscosity predicted value Y 门尼预测 and the preset Mooney expected value Y 门尼期望 By comparing and calculating the absolute deviation, it can effectively ensure that the process parameters in the actual rubber mixing process meet the expected settings, ensure the effect of rubber mixing, and reduce adverse effects and errors.
[0112] Specifically, step S100 of this embodiment includes:
[0113] S110, collect data and create data set X 过程历史 and Y 门尼历史 ;
[0114] S120, the data set X obtained in step S100 过程历史 and Y 门尼历史 Perform data cleaning and obtain a normal data set X 过程正常 and Y 门尼正常 ;
[0115] S130: the data set X obtained in step S120 过程正常 and Y 门尼正常 Genetic programming to generate new features X 过程特征合成 , and added to the original X 过程正常 To generate a new dataset X 过程候选 ;
[0116] S140, the data set X obtained in step S130 过程候选 The target Mooney viscosity Y is selected by the maximum correlation minimum redundancy algorithm. 门尼历史 The most relevant input feature X corr ;
[0117] S150: the input feature X obtained in step S140 corr and Y 门尼正常 Use extreme gradient boosting algorithm to achieve regression prediction and establish model Mooney prediction model 门尼预测 .
[0118] Preferably, the step S110 is specifically as follows:
[0119] Collect parameters during the mixing process to form a multivariate parameter data set X 过程历史 , parameters include but are not limited to: current, temperature, pressure, speed, energy, power and other multivariate parameter data sets X 过程历史 , define its corresponding independent variable X 样本 =(x1, x2, ..., x m ) T , where m is the number of parameters, and the entire data set X 过程历史 Then it is defined as X 过程历史 =(X 样本1 , X 样本2 , …, X 样本n ) where n is the number of cars in historical data; Mooney viscosity value corresponds to X 过程历史 , Y 门尼历史 =(y1, y2, ..., y n ) T , where n is the number of cars in the historical data; 门尼历史 Compared with the actual mixing process data X 过程历史 Align and complete the mapping of data labels and inputs.
[0120] Preferably, the step S120 is specifically as follows:
[0121] By using the Isolation Forest algorithm IsolationForest to remove X 过程历史 Neutralize Y 门尼历史 outliers in the dataset X to obtain the normal 过程正常 and Y 门尼正常 ;
[0122] By constructing multiple Isolation Trees and calculating the average path length of each tree, the final anomaly score is obtained.
[0123] For a data point x, its path length in the tree is recorded as h(x);
[0124] The expected average path length can be approximated by the following formula:
[0125]
[0126] Where H(i) is the i-th harmonic number:
[0127]
[0128] Based on the comparison of path length with the average path length, an anomaly score can be defined:
[0129]
[0130] Among them, s(x, n) close to 1 indicates that the data point x is an outlier, and close to 0 indicates a normal point.
[0131] Preferably, the S130 is specifically to generate a new expression by combining existing features and operators, and select an expression with higher fitness as a new feature, wherein the fitness is the mixing process X 过程正常 Generate Pearson correlation coefficient between features and targets;
[0132] include
[0133] A. Initial population generation: define an operator set O and a terminator set T (variables and constants), randomly generate several expression trees, and the nodes of each tree are randomly selected from O and T;
[0134] B. Fitness evaluation, using the Pearson correlation coefficient to measure the linear correlation between the generated features and the target variable:
[0135]
[0136] Among them, x i is the generated eigenvalue, y i is the target variable value, and They are x i and i The fitness function can be defined as the absolute value of the correlation coefficient
[0137] f(x)=|r|
[0138] The higher the coefficient of correlation, the higher the fitness;
[0139] C. Selection: The selection operation determines which individuals will be used to generate the next generation based on their fitness. The probability of an individual being selected is proportional to its fitness, and the probability of selection pi can be expressed as:
[0140]
[0141] D. Crossover operation: The crossover operation exchanges part of the expression trees of two parent individuals to generate new offspring. The steps of the crossover operation are as follows:
[0142] Randomly select two parent individuals (A) and (B);
[0143] Randomly select a subtree from (A) and (B);
[0144] Interact these two subtrees to generate two new individuals;
[0145] E. Mutation operation: Mutation operation makes small-scale random modifications to individuals to increase the diversity of the population. The steps of mutation operation are as follows:
[0146] Randomly select an individual;
[0147] Randomly select a subtree from the individual;
[0148] Replace the subtree with a randomly generated subtree;
[0149] F. Termination: When any of the following termination conditions are met, the termination conditions include:
[0150] The preset maximum number of generations has been reached.
[0151] Find individuals whose fitness meets the characteristic number and exceeds the fitness threshold (0.9).
[0152] Fitness has not improved significantly over several generations;
[0153] G. Select features and select several expressions with the best performance as new features X according to fitness. 过程特征合成 , added to the original X 过程正常 To generate a new dataset X 过程候选 .
[0154] Preferably, the step S140 is specifically to obtain the input feature X through the following steps: corr ,include
[0155] a. Relevance, specifically
[0156] Use mutual information (MI) to define the relationship between the correlation measurement feature and the target variable, and define feature X i and the target variable Y, whose mutual information is defined as
[0157]
[0158] Among them, (p(x i ,y)) is (X i ) and (Y), (p(x i )) and (p(y)) are respectively (X i ) and (Y);
[0159] b. Redundancy, specifically
[0160] Mutual information is used to quantify the redundant information between redundancy measurement features. Given two features Xi and Xj, their redundant mutual information is defined as:
[0161]
[0162] Among them, p(x i ,x j ) is X i and X j The joint probability distribution of i ) and p(x j ) are X i and X j The marginal probability distribution of
[0163] c. mRMR objective function
[0164] Select a subset of features so that the correlation between these features and the target variable is maximized and the redundancy between these features is minimized. The objective function is defined as:
[0165]
[0166] Among them, S is the selected feature subset, |S| is the size of the subset;
[0167] d. Stepwise selection process
[0168] Initialization: Select the first feature so that it has the greatest correlation with the target variable Y;
[0169] Iteration selection:
[0170] At each step, a new feature Xk is selected so that the correlation between the feature and the target variable Y minus the redundancy between the feature and the currently selected feature subset is maximized, that is, the feature Xk selected at each step satisfies:
[0171]
[0172] Among them, Sk-1 is the feature subset that has been selected in the previous k-1 steps;
[0173] e. Termination conditions
[0174] The feature selection process usually terminates when the preset number of features or other stopping conditions are reached, and several variables are selected as the final modeling data set Xcorr.
[0175] Preferably, the S150 specifically includes
[0176] (1) Given a training data set X corr , Y 门尼正常 ;
[0177] (2) Initialize the model f0(x) to make it equal to the mean of all samples;
[0178] (3) For each iteration t = 1, 2, ..., T,
[0179] Calculate the negative gradient of the loss function as the residual of the current model, i.e., the pseudo residual
[0180] Use pseudo residuals to train a new decision tree and get the output value of the leaf node
[0181] Update the model Where η is the learning rate;
[0182] After completing the update iteration, the model Mooney prediction model Model is obtained 门尼预测 .
[0184] The above-described embodiments only express a certain implementation mode of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the patent scope of the present invention. It should be pointed out that, for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.
Claims
1. A method for soft measurement and process control of Mooney viscosity in a mixing process of an internal mixer, characterized in that: The method comprises the following steps: S100, establish Mooney online prediction model Model 门尼预测 ; S200, real-time data X 过程实时 Input to the Mooney online prediction model Model established in step S100 门尼预测 Calculate the predicted Mooney viscosity value Y of the current mixing process 门尼预测 ; S300, the predicted Mooney viscosity value Y obtained in step S200 门尼预测 and the preset Mooney expected value Y 门尼期望 Compare and calculate the absolute deviation. If the deviation is within the preset error range Err 范围 If the deviation is not within the preset error range Err, then step S400 is executed. 范围 If the number of steps is within 1, step S200 is repeated; S400, send mixing process end sequence signal 结束 Control the internal mixer PLC to complete the mixing end operation.
2. The method for soft measurement and process control of Mooney viscosity in a mixing process of an internal mixer according to claim 1, characterized in that: The step S100 specifically includes S110, collect data and create data set X 过程历史 and Y 门尼历史 ; S120, the data set X obtained in step S100 过程历史 and Y 门尼历史 Perform data cleaning and obtain a normal data set X 过程正常 and Y 门尼正常 ; S130, the data set X obtained in step S120 过程正常 and Y 门尼正常 Genetic programming to generate new features X 过程特征合成 , and added to the original X 过程正常 To generate a new dataset X 过程候选 ; S1 40. The data set X obtained in step S130 过程候选 The target Mooney viscosity Y is selected by the maximum correlation minimum redundancy algorithm. 门尼历史 The most relevant input feature X corr ; S150: the input feature X obtained in step S140 corr and Y 门尼正常 Use extreme gradient boosting algorithm to achieve regression prediction and establish model Mooney prediction model 门尼预测 .
3. The method for soft measurement and process control of Mooney viscosity in a mixing process of an internal mixer according to claim 2, characterized in that: The step S110 is specifically as follows: Collect parameters during the mixing process to form a multivariate parameter data set X 过程历史 , define its corresponding independent variable X 样本 =(x1, x2, ..., x m ) T , where m is the number of parameters, and the entire data set X 过程历史 Then it is defined as X 过程历史 =(X 样本1 , X 样本2 , …, X 样本n ) where n is the number of cars in historical data; Mooney viscosity value corresponds to X 过程历史 , Y 门尼历史 =(y1, y2, ..., y n ) T , where n is the number of cars in the historical data; 门尼历史 Compared with the actual mixing process data X 过程历史 Align and complete the mapping of data labels and inputs.
4. The method for soft measurement and process control of Mooney viscosity in a mixing process of an internal mixer according to claim 3, characterized in that: The step S120 is specifically as follows: By using the Isolation Forest algorithm IsolationForest to remove X 过程历史 Neutralize Y 门尼历史 outliers in the dataset X to obtain the normal 过程正常 and Y 门尼正常 ; By constructing multiple Isolation Trees and calculating the average path length of each tree, the final anomaly score is obtained. For a data point x, its path length in the tree is recorded as h(x); The expected average path length can be approximated by the following formula: Where H(i) is the i-th harmonic number: Based on the comparison of path length with the average path length, an anomaly score can be defined: Among them, s(x, n) close to 1 indicates that the data point x is an outlier, and close to 0 indicates a normal point.
5. The method for soft measurement and process control of Mooney viscosity in a mixing process of an internal mixer according to claim 4, characterized in that: The S130 specifically generates a new expression by combining existing features and operators, and selects an expression with a higher fitness as a new feature, wherein the fitness is the mixing process X 过程正常 Generate Pearson correlation coefficient between features and targets; include A. Initial population generation: define an operator set O and a terminator set T (variables and constants), randomly generate several expression trees, and the nodes of each tree are randomly selected from O and T; B. Fitness evaluation, using the Pearson correlation coefficient to measure the linear correlation between the generated features and the target variable: Among them, x i is the generated eigenvalue, y i is the target variable value, and They are x i and i The fitness function can be defined as the absolute value of the correlation coefficient. f(x)=|r| The higher the coefficient of correlation, the higher the fitness; C. Selection: The selection operation determines which individuals will be used to generate the next generation based on their fitness. The probability of an individual being selected is proportional to its fitness, and the probability of selection pi can be expressed as: D. Crossover operation: The crossover operation exchanges part of the expression trees of two parent individuals to generate new offspring. The steps of the crossover operation are as follows: Randomly select two parent individuals (A) and (B); Randomly select a subtree from (A) and (B); Interact these two subtrees to generate two new individuals; E. Mutation operation: Mutation operation makes small-scale random modifications to individuals to increase the diversity of the population. The steps of mutation operation are as follows: Randomly select an individual; Randomly select a subtree from the individual; Replace the subtree with a randomly generated subtree; F. Termination: When any of the following termination conditions are met, the termination conditions include: Reach the preset maximum number of generations; Find individuals whose fitness meets the characteristic number and exceeds the fitness threshold (0.9); Fitness has not improved significantly over several generations; G. Select features and select several expressions with the best performance as new features X according to fitness. 过程特征合成 , added to the original X 过程正常 To generate a new dataset X 过程候选 .
6. The method for soft measurement and process control of Mooney viscosity in a mixing process of an internal mixer according to claim 5, characterized in that: S140 specifically obtains the input feature X through the following steps corr ,include a. Relevance, specifically Use mutual information (MI) to define the relationship between the correlation measurement feature and the target variable, and define feature X i and the target variable Y, whose mutual information is defined as Among them, (p(x i ,y)) is (X i ) and (Y), (p(x i )) and (p(y)) are respectively (X i ) and (Y); b. Redundancy, specifically Mutual information is used to quantify the redundant information between redundancy measurement features. Given two features Xi and Xj, their redundant mutual information is defined as: Among them, p(x i ,x j ) is X i and X j The joint probability distribution of i ) and p(x j ) are X i and X j The marginal probability distribution of c. mRMR objective function Select a subset of features so that the correlation between these features and the target variable is maximized and the redundancy between these features is minimized. The objective function is defined as: Among them, S is the selected feature subset, |S| is the size of the subset; d. Stepwise selection process Initialization: Select the first feature so that it has the greatest correlation with the target variable Y; Iteration Selection: At each step, a new feature Xk is selected so that the correlation between the feature and the target variable Y minus the redundancy between the feature and the currently selected feature subset is maximized, that is, the feature Xk selected at each step satisfies: Among them, Sk-1 is the feature subset that has been selected in the previous k-1 steps; e. Termination conditions The feature selection process usually terminates when the preset number of features or other stopping conditions are reached, and several variables are selected as the final modeling data set Xcorr.
7. The method for soft measurement and process control of Mooney viscosity in a mixing process of an internal mixer according to claim 6, characterized in that: The S150 specifically includes (1) Given a training data set X corr , Y 门尼正常 ; (2) Initialize the model f0(x) to make it equal to the mean of all samples; (3) For each iteration t = 1, 2, ..., T, Calculate the negative gradient of the loss function as the residual of the current model, i.e., the pseudo residual Use pseudo residuals to train a new decision tree and get the output value of the leaf node Update the model Where η is the learning rate; After completing the update iteration, the model Mooney prediction model Model is obtained 门尼预测 .
