A multi-objective multivariable intelligent control method for a grinding process

Abstract

This invention discloses a multi-objective, multi-variable intelligent control method for the refining process, comprising: describing the refining process using a bilinear model; when new quality variable data is available and the amount of quality variable data is less than the moving window size, rapidly learning the gain parameters of the bilinear model based on recursive least squares; when new quality variable data is available and the current amount of quality variable data is greater than the moving window size, rapidly learning the gain parameters of the bilinear model based on the moving window and batch least squares; updating the set power of the refining motor based on the gain coefficient matrix, the bilinear model, and the quality variable deviation; when new process variables are available, calculating the correction coefficient matrix of the bilinear model using recursive least squares based on historical data and the gain matrix of the bilinear model; and updating the set power of the refining motor based on the gain coefficient matrix, the correction coefficient matrix, the bilinear model, and the quality variable deviation. This invention achieves real-time optimized control of the refining process.

Description

Technical Field

[0001] This invention relates to the technical field of pulping process control, and in particular to a multi-objective, multi-variable intelligent control method for the pulping process. Background Technology

[0002] In pulping, freeness (SR) and fiber length (FL) are core quality indicators for controlling the pulping process, directly affecting the strength, air permeability, and uniformity of paper. These two indicators are closely related to the pulp concentration at the pulp inlet, pulp flow rate, pulping process temperature, pressure, and pulping motor power. Among these, controlling the pulping motor power is the most important means of controlling pulping quality indicators.

[0003] Most existing pulping process control systems employ distributed control systems (DCS), while the set power of the pulping motor is still adjusted manually. Specifically, operators periodically sample the pulp at the outlet, measure the freeness and fiber length in a laboratory, and then adjust the set power of the pulping motor based on their experience via a computer interface. Finally, the power control system of the pulping motor completes the power adjustment. However, manually adjusting the set power of the pulping motor still relies on manual experience, which presents the following problems: 1) It only considers the impact of the motor power on pulp quality, neglecting the influence of pulp concentration and flow rate at the pulp inlet. Considering these factors manually is very complex, and operators cannot calculate them accurately; 2) It heavily relies on the operator's experience and skill level, leading to significant fluctuations in pulp quality among different operators; 3) It cannot adapt to the wear of the mill itself or automatically correct for wear-related power adjustments; 4) It lacks adaptability to different production conditions and pulp formulations, potentially leading to significant batch-to-batch variations in pulp quality indicators.

[0004] Therefore, by using optimized control methods to automatically adjust the set power of the refining motor based on the automatic sampling and measurement results of fiber length and beating degree, the labor intensity of operators can be greatly reduced and the refining quality can be improved. At the same time, the influence of pulp concentration and pulp flow fluctuations on the refining quality can also be eliminated. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings and deficiencies of the prior art and provide a multi-objective and multi-variable intelligent control method for the pulping process. This method establishes a multi-objective and multi-variable bilinear model to describe the pulping process, and provides an adaptive machine learning algorithm for the model parameters and a control algorithm for the set power of the pulping motor. The above-mentioned bilinear model and algorithm can achieve real-time optimized control of the pulping process under dynamic conditions such as pulp concentration fluctuation, pulp flow fluctuation, and wear changes of the grinding disc teeth.

[0006] To achieve the above objectives, the technical solution provided by this invention is as follows: a multi-objective, multi-variable intelligent control method for the pulping process. This method is a pulp quality variable control method based on soft sensing, fast machine learning, and incremental control algorithms. The method uses a bilinear model to describe the pulping process. This bilinear model takes process variables as input variables and quality variables as output variables. The process variables include the power of the pulping motor, the pulp flow rate at the pulp inlet, and the pulp concentration. The quality variables include the freeness at the pulp outlet and the fiber length. The machine learning algorithm is used to adjust the bilinear model parameters according to the quality variables. The soft sensing is used to dynamically fine-tune the bilinear model parameters according to the process variables. The incremental control algorithm calculates the adjustment value of the set power of each pulping motor based on the adjusted bilinear model and the deviation of the quality variables.

[0007] The multi-objective, multi-variable intelligent control method is divided into a machine learning stage and a soft measurement stage based on the different sampled data, i.e., the collected data samples.

[0008] After sampling quality variables, a machine learning step is performed, which includes the following steps:

[0009] 1.1) Adjust the parameters of the bilinear model based on the quality variable. When the sample size is small, the recursive least squares method is used for parameter adjustment. When the sample size is large, the batch least squares method based on the moving window is used for parameter adjustment.

[0010] 1.2) Adjust the set power of the grinding motor based on the updated bilinear model and incremental control algorithm;

[0011] 1.3) Constrain the adjustment value of the set power of the refining motor and update the set power of the refining motor according to the parameters of the refining motor, wherein the parameters of the refining motor include the resolution of the set power of the refining motor and its upper and lower limits;

[0012] After sampling process variables, a soft measurement step is performed, which includes the following steps:

[0013] 2.1) Fine-tuning the bilinear model parameters based on process variables: Specifically, based on the gain coefficients calculated after the previous mass variable sampling and historical data, the correction coefficients related to pulp concentration and pulp flow rate in the bilinear model are calculated using the recursive least squares method. Then, the gain of the pulping motor power on the mass variables is corrected based on the pulp concentration and pulp flow rate values ​​in the process variables, and the bilinear model parameters are updated.

[0014] 2.2) The adjustment value of the set power of the grinding motor is obtained based on the updated bilinear model and incremental control algorithm;

[0015] 2.3) The set power of the grinding motor is adjusted based on the set power resolution and its upper and lower limits, and the set power of the grinding motor is updated.

[0016] Furthermore, in the initialization stage of the pulping process, a bilinear model is established to describe the pulping process, and the specific formula is as follows:

[0017] ;

[0018] In the formula, For pulp concentration, For slurry flow rate, It is the product of slurry concentration and slurry flow rate. This is the average value of the product of slurry concentration and slurry flow rate; The number of refining motors connected in series in the refining process; power. For the first Net power of the pulping motor, which is the power of the pulping motor minus the no-load power; and The first and second cases, respectively, are when the effects of concentration and flow rate are not considered. The gain coefficient of net power of the grinding motor on the degree of beating. and The first and second cases, respectively, are when the effects of concentration and flow rate are not considered. The gain coefficient of the net power of the pulping motor on fiber length. and These are the initial freeness and initial fiber length of the pulp, respectively. and These represent the effects of slurry concentration and slurry flow rate on the concentration and flow rate, respectively. and Correction factor; and These are the freeness and fiber length at the pulp outlet, respectively;

[0019] The matrix form of the bilinear model is as follows:

[0020] ;

[0021] In the formula, The number of data measurements refers to the number of measurements taken, including measurements of pulp quality variables and process variables; this will be used in the following text. Indicates the first The total number of measurements corresponding to each pulping quality variable measurement. In the Next to The values ​​taken between the measurements of the quality variables of the secondary refining process are: ,in It is the measurement of process variables between quality variable sampling; , They are the first The first and second mass variable values ​​during the second data measurement are the freeness and fiber length at the pulp outlet. It is the first The mass variable vector during the next data measurement; and They are the first Initial freeness and initial fiber length at the time of this data measurement; It is the first Initial value vector of quality variables during the next data measurement; and They are the first The first data measurement The gain coefficient of the net power of the pulping motor on the degree of beating and fiber length; and They are the first Gain coefficient matrix of the net power of the refining motor to the mass variable, with and without considering the effects of pulp concentration and pulp flow rate during the second data measurement; and They are the first The first data measurement When considering the effects of pulp concentration and pulp flow rate on the refining motor, the pulp concentration and pulp flow rate have a significant impact. and Correction factor; When considering the effects of pulp concentration and pulp flow rate, and The correction coefficient matrix; It is the first The first data measurement Net power of the grinding motor; It is the first Net power vector of the grinding motor during the next data measurement; It is the first The first data measurement Net set power of the grinding motor; It is the net set power vector of the grinding motor.

[0022] Furthermore, in step 1.1), when there are new data samples in the grinding process and the sample size is less than the moving window size, the recursive least squares method is used to adjust the parameters of the bilinear model. When there are new data samples and the sample size is greater than the moving window size, the batch least squares method based on the moving window is used to adjust the parameters of the bilinear model. That is, the parameters of the bilinear model will be updated every time there are new data samples.

[0023] Furthermore, in step 1.2), based on the updated bilinear model and the mass variable deviation in the samples, an incremental control algorithm is used to calculate the adjustment value of the set power of the grinding motor that can eliminate the mass variable deviation, as shown in the following formula:

[0024] ;

[0025] In the formula, , These are the target values ​​for the first and second pulp quality variables, respectively. , The first The measured values ​​of the first and second polishing quality variables during the second polishing quality variable measurement; , The first The deviations between the first and second measured values ​​of the pulp quality variables and the target value; For the first The deviation vector between the measured values ​​and the target values ​​of the secondary pulp quality variables; For the first The gain coefficient matrix of the net power of the refining motor on the mass variables is used when measuring the mass variables of the refining motor without considering the effects of pulp concentration and pulp flow rate. The robustness coefficient; For unit array; For the first The power increment vector of the grinding motor setting for the secondary grinding quality variable measurement data; It is the first Net power vector of the grinding motor during secondary grinding quality variable measurement; It is the first Net set power of the grinding motor during the first data measurement;

[0026] To avoid setpoint oscillations and improve steady-state accuracy, the following formula is used:

[0027] ;

[0028] In the formula, This is the maximum power increment set for the pulping motor; For unit array; Based on the The original set power adjustment value vector of the grinding motor obtained by measuring and calculating the grinding quality variable; Based on the The incremental power vector of the grinding motor after incremental constraint is obtained by calculating the measurement data of the grinding quality variable. It is the first The net set power of the pulping motor at the time of the data measurement is based on the first... The net set power of the grinding motor updated by the next grinding quality variable measurement data, plus the no-load power, is used as the final updated power set value of the grinding motor.

[0029] When the process requires the set power of the i-th refining motor to remain unchanged, and only the set power of other refining motors is adjusted, the calculation formula is:

[0030] ;

[0031] In the formula, It is the first The gain coefficient matrix of the net power of the refining motor on the mass variables is used when measuring the mass variables of the refining motor without considering the effects of pulp concentration and pulp flow rate. It is the first The gain matrix of the grinding motor power on the quality variable is obtained by removing the relevant components of the i-th grinding motor when measuring the grinding quality variable. It is the first When measuring the quality variables of the next grinding process, the grinding motor power vector with the relevant components of the i-th grinding motor removed; For unit array; It is the first The power adjustment value vector of the i-th grinding motor is removed when measuring the grinding quality variable. Based on the The net set power vector of the grinding motor is updated by measuring the grinding quality variable and removing the relevant components of the i-th grinding motor.

[0032] Further, in step 2.1), based on the mean of the product of the average pulp concentration and pulp flow rate over a recent period and the average pulp concentration and pulp flow rate of historical samples, the gain coefficient of the refining motor power on the mass variable in the bilinear model is adjusted; when mass variable data is available, the gain matrix is ​​updated. After that, consider and The effect is calculated by applying parameter machine learning algorithms. and That is, the correction coefficient matrix The matrix form formula is as follows:

[0033] ;

[0034] ;

[0035] In the formula, It is the first The product of pulp concentration and pulp flow rate when measuring the quality variables of the pulping process; The measurement data for the pulp quality variable is less than the size of the moving window. The situation at that time; For the measurement data of the pulp quality variable to be greater than or equal to the moving window size The situation at that time; It is the first The mean of the product of pulp concentration and pulp flow rate during the measurement of pulp quality variables in each refining step; It is the first Net power vector of the grinding motor during secondary grinding quality variable measurement; It is a unit array; , They are the first sequence Gain auxiliary matrix for measuring secondary pulping quality variables; It is the first Kalman gain matrix for measuring the quality variables of the secondary grinding process; It is the first Intermediate variable matrix for measuring the quality variables of the second refining process; It is the first Initial value vector of mass variables during the measurement of mass variables in the second grinding process; When the effects of pulp concentration and pulp flow rate are not considered, the first The gain matrix of the net power of the grinding motor on the mass variable during the measurement of the mass variable of the grinding motor; It is the first Mass variable vector during secondary grinding quality variable measurement; , They are the first sequence When considering the effects of pulp concentration and flow rate on the quality variables of secondary refining, the influence of pulp concentration and flow rate on... and The correction coefficient matrix; It is the first When considering the effects of pulp concentration and flow rate in the measurement of secondary pulping quality variables, the influence of pulp concentration and flow rate on... and The adjustment matrix of correction coefficients; Forgetting factor; The robustness coefficient; The initial value for iteration can take a default value. It is a unit array; initial value .

[0036] Furthermore, in step 2.2), after obtaining the gain matrix... and correction coefficient matrix Then, calculate the new set power of the pulping motor. The calculation formula is as follows:

[0037] ;

[0038] Due to the variable of slurry concentration in the process Mixing flow rate The sampling period is shorter than the control period. , Represents all data within the data window The average value, and They are the first The first time the quality variable of the pulping was measured. When considering the effects of pulp concentration and pulp flow rate on the refining motor, the pulp concentration and pulp flow rate have a significant impact. and Correction factor; It is the first The net set power of the pulping motor is updated during this data measurement.

[0039] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0040] 1. In the initial stage, a parametric machine learning algorithm based on recursive least squares was adopted, which shortened the time to reach the optimization working point and reduced the amount of waste slurry.

[0041] 2. After multiple samplings, a parameter machine learning algorithm based on batch least squares with a moving window is adopted, which improves the ability to resist random interference. The sample data before the moving window is removed, which can better handle the time-varying situation caused by gear wear and improve the control performance.

[0042] 3. Soft measurement was performed between two pulp quality samplings, which can estimate the impact of concentration and flow rate changes on pulp quality variables during the pulping process, and can adjust the milling power in a timely manner, thus improving the real-time performance of optimized control.

[0043] 4. It enables automatic control of the pulping process, resulting in more stable pulping quality compared to manual control.

[0044] 5. It can adapt to different production conditions and pulp formulations, and can significantly reduce batch-to-batch variability. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of a pulping production line. In the diagram, mills 1, 2, 3, 4, and 5 represent the first, second, third, fourth, and fifth pulping motors, respectively.

[0046] Figure 2 Optimize the overall control flow chart for the pulping process. Detailed Implementation

[0047] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0048] This embodiment uses Taking a pulping production line consisting of 5 pulping motors as an example, this invention provides a detailed explanation of a multi-objective, multi-variable intelligent control method for the pulping process. The layout of this production line is as follows: Figure 1 As shown.

[0049] The pulping control system consists of a distributed control system (DCS) and monitoring software. The DCS monitors raw material concentration, flow rate, and actual power of the refiner motor in real time, automatically samples pulp, measures freeness and fiber length, and executes the refiner motor power settings configured by the monitoring software. The monitoring software periodically acquires and stores quality and process variable data from the DCS, and updates the refiner motor power settings using intelligent control methods. The control flow of the pulping control system is as follows: Figure 2 As shown.

[0050] Step 201, Initialization of the optimization control of the pulping process: The operator selects the specific pulping process through the human-machine interface, which mainly includes the number of pulping motors, the working status of each pulping motor, the power setting range, the control cycle, and the power resolution.

[0051] The pulping control system determines the bilinear model of the pulping process based on the selected process, as shown in Equation (1) below, and initializes the model parameters and the set power of the pulping motor with missing values ​​or the final parameter values ​​under the same process conditions.

[0052] (1);

[0053] In the formula, For pulp concentration, For slurry flow rate, It is the product of slurry concentration and slurry flow rate. This is the average value of the product of slurry concentration and slurry flow rate; The number of refining motors connected in series in the refining process; For the first Net power of the pulping motor, which is the power of the pulping motor minus the no-load power; and The first and second cases, respectively, are when the effects of concentration and flow rate are not considered. The gain coefficient of net power of the grinding motor on the degree of beating. and The first and second cases, respectively, are when the effects of concentration and flow rate are not considered. The gain coefficient of the net power of the pulping motor on fiber length. and These are the initial freeness and initial fiber length of the pulp, respectively. and These represent the effects of slurry concentration and slurry flow rate on the concentration and flow rate, respectively. and Correction factor; and These are the freeness and fiber length at the pulp outlet, respectively;

[0054] The matrix form of the bilinear model is as follows:

[0055] (2);

[0056] In the formula, The number of data measurements refers to the number of measurements taken, including measurements of pulp quality variables and process variables; this will be used in the following text. Indicates the first The total number of measurements corresponding to each pulping quality variable measurement. In the Next to The values ​​taken between the measurements of the quality variables of the secondary refining process are: ,in It is the measurement of process variables between quality variable sampling; , They are the first The first and second mass variable values ​​during the second data measurement are the freeness and fiber length at the pulp outlet. It is the first The mass variable vector during the next data measurement; and They are the first Initial freeness and initial fiber length at the time of this data measurement; It is the first Initial value vector of quality variables during the next data measurement; and They are the first The first data measurement The gain coefficient of the net power of the pulping motor on the degree of beating and fiber length; and They are the first Gain coefficient matrix of the net power of the refining motor to the mass variable, with and without considering the effects of pulp concentration and pulp flow rate during the second data measurement; and They are the first The first data measurement When considering the effects of pulp concentration and pulp flow rate on the refining motor, the pulp concentration and pulp flow rate have a significant impact. and Correction factor; When considering the effects of pulp concentration and pulp flow rate, and The correction coefficient matrix; It is the first The first data measurement Net power of the grinding motor; It is the first Net power vector of the grinding motor during the next data measurement; It is the first The first data measurement Net set power of the grinding motor; It is the first The net set power vector of the grinding motor during this data measurement.

[0057] Step 202, Quality Data Quantity Judgment: When the amount of pulp quality variable data is less than the size of the moving window... At this point, it enters the machine learning phase based on recursive least squares; reaching the point where the window size is adjusted. Then it enters the machine learning stage based on moving windows and batch least squares.

[0058] Steps 210 and 220, Data Type Determination: Based on the type of new data, different control steps are initiated. Quality variable samples are used to update model parameters and adjust mill power; process variable samples are used for soft measurement and fine-tuning of the grinding power.

[0059] Step 211, Machine learning step based on recursive least squares: Update model parameters based on new quality variable samples.

[0060] Without considering the effects of slurry concentration and flow rate, calculate the model parameters. :

[0061] (3);

[0062] In the formula, It is the first The vector of mass variable measurement values ​​obtained from the second mass variable measurement; As a unit array, Forgetting factor; It is the first Net power vector of the grinding motor during secondary grinding quality variable measurement; When the effects of concentration and flow rate are not considered Gain auxiliary matrix for secondary mass variable measurement; When the effects of concentration and flow rate are not considered Kalman gain vector during sub-mass variable measurement; It is the first Initial value vector of mass variables during the second mass variable measurement; It is the first When considering the effects of pulp concentration and flow rate on the quality variables of secondary refining, the influence of pulp concentration and flow rate on... and The correction coefficient matrix; The robustness coefficient. During the rapid learning phase, remain unchanged. initial value .

[0063] Then, a parametric machine learning algorithm is applied to calculate the correction coefficient matrix. :

[0064] (4);

[0065] In the formula, It is the first The next quality variable measurement is the average of the slurry concentration and slurry flow rate at the slurry inlet during the previous control period; It is the first The product of pulp concentration and pulp flow rate, measured as secondary mass variables; It is the first The mean of the product of slurry concentration and slurry flow rate, measured as secondary mass variables; It is the first Gain auxiliary matrix for secondary mass variable measurement; It is the first Kalman gain vector for secondary mass variable measurement; It is the first The intermediate variable matrix for secondary quality variable measurements; When the effects of pulp concentration and pulp flow rate are not considered, the first The gain matrix of the net power of the grinding motor on the mass variable during the measurement of the mass variable of the grinding motor; It is the first When considering the effects of pulp concentration and flow rate on the quality variables of secondary refining, the influence of pulp concentration and flow rate on... and The correction coefficient matrix; Forgetting factor; The robustness coefficient; The initial value can be a default value. Initial value Correction coefficient matrix Used for subsequent soft measurement steps, which are based on the gain matrix. and correction coefficient matrix The gain matrix considering the effects of slurry concentration and slurry flow rate is obtained. .

[0066] Then, the set power of the pulping motor is updated using an incremental control algorithm:

[0067] (5);

[0068] In the formula, , These are the target values ​​for the first and second pulp quality variables, respectively. , The first The measured values ​​of the first and second polishing quality variables during the second polishing quality variable measurement; , The first The deviations between the first and second measured values ​​of the pulp quality variables and the target value; For the first The deviation vector between the measured values ​​and the target values ​​of the secondary pulp quality variables; For the first The gain coefficient vector of the net power of the refining motor on the mass variable is not considered when measuring the mass variable of the refining motor, which does not take into account the effects of pulp concentration and pulp flow rate. The robustness coefficient; For unit array; For the first The power increment vector of the grinding motor setting for the secondary grinding quality variable measurement data; It is the first Net power vector of the grinding motor during secondary grinding quality variable measurement; It is the first Net set power of the grinding motor during the first data measurement;

[0069] When the process requires the grinding motor setting power of mill i to remain unchanged, the power setting values ​​of other mills are adjusted as shown in the following formula (6):

[0070] (6);

[0071] In the formula, It is the first The gain coefficient matrix of the net power of the refining motor on the mass variables is used when measuring the mass variables of the refining motor without considering the effects of pulp concentration and pulp flow rate. It is the first The gain matrix of the grinding motor power on the quality variable is obtained by removing the relevant components of the i-th grinding motor when measuring the grinding quality variable. It is the first When measuring the quality variables of the next grinding process, the grinding motor power vector with the relevant components of the i-th grinding motor removed; For unit array; It is the first The power adjustment value vector of the i-th grinding motor is removed when measuring the grinding quality variable. Based on the The net set power vector of the grinding motor is updated by measuring the grinding quality variable and removing the relevant components of the i-th grinding motor.

[0072] Step 212, Soft Sensing and Power Fine-Tuning: Based on the calculated input of new process variable data and gain matrix... Correction coefficient matrix Using a bilinear model, a model parameter matrix considering the effects of slurry concentration and slurry flow rate is obtained. The components:

[0073] (7);

[0074] Based on the gain matrix considering the effects of slurry concentration and slurry flow rate Calculate the new set power of the pulping motor using the following formula:

[0075] (8);

[0076] Step 221, Machine learning based on moving window and batch least squares: When new high-quality data samples are available, the latest ones are retrieved. A moving window is formed by a sample of quality data. The next iteration yields the model parameters without considering the effects of slurry concentration and slurry flow rate. :

[0077] (9);

[0078] In the formula, The number of samples for the moving window;

[0079] Then, based on the data in the moving window, a parametric machine learning algorithm is applied to update the correction coefficient matrix. :

[0080] (10);

[0081] Correction coefficient matrix Used for subsequent soft measurement steps, which are based on the gain matrix. and correction coefficient matrix The gain matrix considering the effects of slurry concentration and slurry flow rate is obtained. .

[0082] Then, the control increment is calculated, taking into account the maximum power increment, to obtain the new mill power setting value:

[0083] (11);

[0084] In the formula, For unit array; This is the maximum power increment set for the pulping motor; It is the original set power adjustment value vector of the grinding motor obtained based on the j-th grinding quality variable measurement; Based on the The power increment vector of the grinding motor is set based on the second grinding quality variable measurement data, which is the first... Vector of adjustable power settings for the grinding motor after incremental constraints for measuring the quality variables of the secondary grinding process; It is the first The net set power of the pulping motor at the time of the data measurement is based on the first... The net set power of the grinding motor updated by the next grinding quality variable measurement data, plus the no-load power, is used as the final updated power set value of the grinding motor.

[0085] When the process requires the power setting of the i-th refining motor to remain unchanged, the method for adjusting the power setting values ​​of other refining motors is as follows:

[0086] (12);

[0087] Step 222, Soft Measurement and Power Fine-Tuning: Based on the calculated input of new process variable data and gain matrix... Correction coefficient matrix Using a bilinear model, a gain matrix considering the effects of slurry concentration and slurry flow rate is obtained. The components:

[0088] (13);

[0089] Based on the gain matrix considering the effects of slurry concentration and slurry flow rate Calculate the new set power of the pulping motor using the following formula:

[0090] (14);

[0091] Step 203, power cutoff quantization: Cut off the set power of the pulping motor within the set range, take an integer multiple of the motor resolution, and add the no-load power of the pulping motor to obtain the final set power, and set it to the distributed control system.

[0092] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A multi-objective multi-variable intelligent control method for a grinding process, characterized in that, This method is a pulp quality variable control method based on soft sensing, fast machine learning, and incremental control algorithms. It employs a bilinear model to describe the pulping process, with process variables as inputs and quality variables as outputs. The process variables include pulping motor power, pulp flow rate at the pulp inlet, and pulp concentration. The quality variables include pulp freeness and fiber length at the pulp outlet. The machine learning algorithm is used to adjust the bilinear model parameters based on the quality variables. The soft sensing is used to dynamically fine-tune the bilinear model parameters based on the process variables. The incremental control algorithm calculates the adjustment values ​​for the set power of each pulping motor based on the adjusted bilinear model and the deviation of the quality variables. The multi-objective, multi-variable intelligent control method is divided into a machine learning stage and a soft measurement stage based on the different sampled data, i.e., the collected data samples. After sampling quality variables, a machine learning step is performed, which includes the following steps: 1.1) Adjust the parameters of the bilinear model based on the quality variable. When the sample size is small, the recursive least squares method is used for parameter adjustment. When the sample size is large, the batch least squares method based on the moving window is used for parameter adjustment. 1.2) Adjust the set power of the grinding motor based on the updated bilinear model and incremental control algorithm; 1.3) Constrain the adjustment value of the set power of the refining motor and update the set power of the refining motor according to the parameters of the refining motor, wherein the parameters of the refining motor include the resolution of the set power of the refining motor and its upper and lower limits; After sampling process variables, a soft measurement step is performed, which includes the following steps: 2.1) Fine-tuning the bilinear model parameters based on process variables: Specifically, based on the gain coefficients calculated after the previous mass variable sampling and historical data, the correction coefficients related to pulp concentration and pulp flow rate in the bilinear model are calculated using the recursive least squares method. Then, the gain of the pulping motor power on the mass variables is corrected based on the pulp concentration and pulp flow rate values ​​in the process variables, and the bilinear model parameters are updated. 2.2) The adjustment value of the set power of the grinding motor is obtained based on the updated bilinear model and incremental control algorithm; 2.3) The set power of the grinding motor is adjusted based on the set power resolution and its upper and lower limits, and the set power of the grinding motor is updated.

2. A multi-objective, multi-variable intelligent control method for a pulp grinding process according to claim 1, characterized in that, In the initial stage of the pulping process, a bilinear model is established to describe the pulping process, and the specific formula is as follows: ； In the formula, For pulp concentration, For slurry flow rate, It is the product of slurry concentration and slurry flow rate. This is the average value of the product of slurry concentration and slurry flow rate; The number of refining motors connected in series in the refining process; power. For the first Net power of the pulping motor, which is the power of the pulping motor minus the no-load power; and The first and second cases, respectively, are when the effects of concentration and flow rate are not considered. The gain coefficient of net power of the grinding motor on the degree of beating. and The first and second cases, respectively, are when the effects of concentration and flow rate are not considered. The gain coefficient of the net power of the pulping motor on fiber length. and These are the initial freeness and initial fiber length of the pulp, respectively. and These represent the effects of slurry concentration and slurry flow rate on the concentration and flow rate, respectively. and Correction factor; and These are the freeness and fiber length at the pulp outlet, respectively; The matrix form of the bilinear model is as follows: ； In the formula, The number of data measurements refers to the number of measurements taken, including measurements of pulp quality variables and process variables; this will be used in the following text. Indicates the first The total number of measurements corresponding to each pulping quality variable measurement. In the Next to The values ​​taken between the measurements of the quality variables of the secondary refining process are: ,in It is the measurement of process variables between quality variable sampling; , They are the first The first and second mass variable values ​​during the second data measurement are the freeness and fiber length at the pulp outlet. It is the first The mass variable vector during the next data measurement; and They are the first Initial freeness and initial fiber length at the time of this data measurement; It is the first Initial value vector of quality variables during the next data measurement; and They are the first The first data measurement The gain coefficient of the net power of the pulping motor on the degree of beating and fiber length; and They are the first Gain coefficient matrix of the net power of the refining motor to the mass variable, with and without considering the effects of pulp concentration and pulp flow rate during the second data measurement; and They are the first The first data measurement When considering the effects of pulp concentration and pulp flow rate on the refining motor, the pulp concentration and pulp flow rate have a significant impact. and Correction factor; When considering the effects of pulp concentration and pulp flow rate, and The correction coefficient matrix; It is the first The first data measurement Net power of the grinding motor; It is the first Net power vector of the grinding motor during the next data measurement; It is the first The first data measurement Net set power of the grinding motor; It is the net set power vector of the grinding motor.

3. The multi-objective, multi-variable intelligent control method for a pulping process according to claim 2, characterized in that, In step 1.1), when there are new data samples in the grinding process and the sample size is less than the moving window size, the recursive least squares method is used to adjust the parameters of the bilinear model. When there are new data samples and the sample size is greater than the moving window size, the batch least squares method based on the moving window is used to adjust the parameters of the bilinear model. That is, the parameters of the bilinear model will be updated every time there are new data samples.

4. The multi-objective, multi-variable intelligent control method for a pulping process according to claim 3, characterized in that, In step 1.2), based on the updated bilinear model and the mass variable deviation in the samples, an incremental control algorithm is used to calculate the adjustment value of the set power of the grinding motor that can eliminate the mass variable deviation, as shown in the following formula: ； In the formula, , These are the target values ​​for the first and second pulp quality variables, respectively. , The first The measured values ​​of the first and second polishing quality variables during the second polishing quality variable measurement; , The first The deviations between the first and second measured values ​​of the pulp quality variables and the target value; For the first The deviation vector between the measured values ​​and the target values ​​of the secondary pulp quality variables; For the first The gain coefficient matrix of the net power of the refining motor on the mass variables is used when measuring the mass variables of the refining motor without considering the effects of pulp concentration and pulp flow rate. The robustness coefficient; For unit array; For the first The power increment vector of the grinding motor setting for the secondary grinding quality variable measurement data; It is the first Net power vector of the grinding motor during secondary grinding quality variable measurement; It is the first Net set power of the grinding motor during the first data measurement; To avoid setpoint oscillations and improve steady-state accuracy, the following formula is used: ； In the formula, This is the maximum power increment set for the pulping motor; For unit array; Based on the The original set power adjustment value vector of the grinding motor obtained by measuring and calculating the grinding quality variable; Based on the The incremental power vector of the grinding motor after incremental constraint is obtained by calculating the measurement data of the grinding quality variable. It is the first The net set power of the pulping motor at the time of the data measurement is based on the first... The net set power of the grinding motor updated by the next grinding quality variable measurement data, plus the no-load power, is used as the final updated power set value of the grinding motor. When the process requires the set power of the i-th refining motor to remain unchanged, and only the set power of other refining motors is adjusted, the calculation formula is: ； In the formula, It is the first The gain coefficient matrix of the net power of the refining motor on the mass variables is used when measuring the mass variables of the refining motor without considering the effects of pulp concentration and pulp flow rate. It is the first The gain matrix of the grinding motor power on the quality variable is obtained by removing the relevant components of the i-th grinding motor when measuring the grinding quality variable. It is the first When measuring the quality variables of the next grinding process, the grinding motor power vector with the relevant components of the i-th grinding motor removed; For unit array; It is the first The power adjustment value vector of the i-th grinding motor is removed when measuring the grinding quality variable. Based on the The net set power vector of the grinding motor is updated by measuring the grinding quality variable and removing the relevant components of the i-th grinding motor.

5. The multi-objective, multi-variable intelligent control method for a pulping process according to claim 4, characterized in that, In step 2.1), based on the mean of the product of the average pulp concentration and flow rate over a recent period and the average pulp concentration and flow rate of historical samples, the gain coefficient of the refining motor power on the mass variable in the bilinear model is adjusted; when mass variable data is available, the gain matrix is ​​updated. After that, consider and The effect is calculated by applying parameter machine learning algorithms. and That is, the correction coefficient matrix The matrix form formula is as follows: ； ； In the formula, It is the first The product of pulp concentration and pulp flow rate when measuring the quality variables of the pulping process; The measurement data for the pulp quality variable is less than the size of the moving window. The situation at that time; For the measurement data of the pulp quality variable to be greater than or equal to the moving window size The situation at that time; It is the first The mean of the product of pulp concentration and pulp flow rate during the measurement of pulp quality variables in each refining step; It is the first Net power vector of the grinding motor during secondary grinding quality variable measurement; It is a unit array; , They are the first sequence Gain auxiliary matrix for measuring secondary pulping quality variables; It is the first Kalman gain matrix for measuring the quality variables of the secondary grinding process; It is the first Intermediate variable matrix for measuring the quality variables of the second refining process; It is the first Initial value vector of mass variables during the measurement of mass variables in the second grinding process; When the effects of pulp concentration and pulp flow rate are not considered, the first The gain matrix of the net power of the grinding motor on the mass variable during the measurement of the mass variable of the grinding motor; It is the first Mass variable vector during secondary grinding quality variable measurement; , They are the first sequence When considering the effects of pulp concentration and flow rate on the quality variables of secondary refining, the influence of pulp concentration and flow rate on... and The correction coefficient matrix; It is the first When considering the effects of pulp concentration and flow rate in the measurement of secondary pulping quality variables, the influence of pulp concentration and flow rate on... and The correction coefficient adjustment value matrix; Forgetting factor; The robustness coefficient; The initial value for iteration can take a default value. It is a unit array; initial value .

6. The multi-objective, multi-variable intelligent control method for a pulping process according to claim 5, characterized in that, In step 2.2), after obtaining the gain matrix... and correction coefficient matrix Then, calculate the new set power of the pulping motor. The calculation formula is as follows: ； Due to the variable of slurry concentration in the process Mixing flow rate The sampling period is shorter than the control period. , Represents all data within the data window The average value, and They are the first The first time the quality variable of the pulping was measured. When considering the effects of pulp concentration and pulp flow rate on the refining motor, the pulp concentration and pulp flow rate have a significant impact. and Correction factor; It is the first The net set power of the pulping motor is updated during the next data measurement.

Images