阅读说明：本技术 一种应用于压力精密控制的自动机器学习系统和方法 (Automatic machine learning system and method applied to pressure precision control ) 是由 杨舒琬 余正涛 何程 方莹 李煜煌 于 2021-03-16 设计创作，主要内容包括：本发明公开了一种应用于压力精密控制的自动机器学习系统和方法,通过自动深度学习,实现了大量程范围内任意多点设定压力值精密控制最优参数的自动生成及控制参数积累,通过监督机制,在大量程范围内任意多点设定压力值控制时选择最优控制参数进行压力精密控制。解决了传统自动压力控制方法存在控制对象工况因素变化引起的不确定性,且控制对象非线性变化造成难以建立精确数学模型等关键难题。实现了在0—100MPa量程范围任意多点压力控制对象±0.005MPa偏差的精密控制。(The invention discloses an automatic machine learning system and method applied to pressure precision control, which realize automatic generation of optimal parameters and control parameter accumulation of any multipoint set pressure value precision control in a wide range through automatic deep learning, and select the optimal control parameters for pressure precision control through a supervision mechanism when the pressure value is controlled at any multipoint set pressure value in the wide range. The method solves the key problems that the traditional automatic pressure control method has uncertainty caused by the change of the working condition factors of the control object, and the control object is difficult to establish an accurate mathematical model due to the nonlinear change of the control object. The precise control of the deviation of +/-0.005 MPa of any multipoint pressure control object in the range of 0-100 MPa is realized.)

1. An automatic machine learning system for precision pressure control, comprising: high accuracy digital pressure sensor, automatic machine learning device, gas pressure precision control device and high-pressure gas-liquid supercharging device connect gradually, wherein:

the high-precision digital pressure sensor is used for measuring pressure and outputting a digital quantity signal;

the automatic machine learning device comprises a data preprocessing module, a monitoring mechanism module, a data set storage module, a deep learning module and a man-machine interaction module which are connected,

the data preprocessing module is used for acquiring output digital quantity signals of the high-pressure gas-liquid supercharging device and removing error data;

the monitoring mechanism module is used for calculating the pressure control deviation pe and the pressure change rate et in real time, calling the optimal parameters in the data set storage module and outputting digital quantity control signals according to the pressure value and the tolerance boundary condition set by a user, and automatically judging and sending a learning instruction to the deep learning module; simultaneously, optimally selecting the multiple learning results, and storing the results into a data set storage module;

the data set storage module is used for storing the historical optimal parameters and the current deep learning optimal selection results and carrying out optimal selection again according to tolerance boundary conditions set by a user;

the deep learning module is used for automatically learning parameters and uploading learning result data to the supervision mechanism module according to the learning instruction sent by the supervision mechanism module, if the learning result is not satisfactory, the deep learning module sends out the learning instruction again, and the deep learning module carries out secondary learning again until an end instruction is received and stops learning;

the human-computer interaction module is used for providing user parameter setting and related parameter display;

the gas pressure precise control device is used for receiving the digital quantity control signal and converting the digital quantity control signal into a precise control gas source signal; and constantly outputting an air source;

the high-pressure gas-liquid supercharging device is used for receiving a precise control gas source signal and linearly outputting a pressure loop.

2. The automatic machine learning system applied to the pressure precise control as claimed in claim 1, wherein the gas pressure precise control device comprises a digital quantity voltage signal device, a high precision proportional pneumatic valve, a pressure stabilizing valve and a gas source; the digital quantity voltage signal device receives a digital quantity control signal through an RS485 communication port and converts the digital quantity control signal into a 0-10VDC direct current output signal, and the signal resolution is 0.0001 VDC; the electric side of the high-precision proportional pneumatic valve receives a 0-10VDC direct current signal and linearly outputs a 0-1.0MPa precision control air source signal; the pressure stabilizing valve is used for outputting the gas output by the gas source with the constant pressure of 1.0MPa and providing the gas to the side of the gas source of the high-precision proportional pneumatic valve.

3. The automatic machine learning system applied to the pressure precise control as claimed in claim 1, wherein the high-pressure gas-liquid pressure boosting device comprises a gas-liquid pressure boosting pump, an output pressure loop and an oil tank; wherein the pressure ratio of the gas-liquid booster pump is 1:100, and the pressure of 0-100 MPa is linearly output to the output pressure loop.

4. An automatic machine learning method applied to pressure precise control is characterized by comprising the following steps:

s1, collecting the pressure of the high-pressure gas-liquid supercharging device for measurement and outputting a digital quantity signal;

s2, calculating a pressure control deviation pe and a pressure change rate et in real time, calling optimal parameters and outputting digital quantity control signals according to a pressure value and tolerance boundary conditions set by a user, and automatically judging whether to send a deep learning instruction or not; during deep learning, optimal selection is carried out on multiple learning results at the same time; and carrying out optimal selection again according to tolerance boundary conditions set by a user;

s3, learning parameters and obtaining learning result data, if the learning result data is not satisfied, sending out the learning instruction again, learning again, and circulating N times until receiving the end instruction to stop learning, and forming a digital quantity control signal;

s4, receiving the digital control signal and converting the digital control signal into a precision control air source signal; and constantly outputting an air source;

and S5, receiving a precision control air source signal and outputting a pressure loop linearly.

5. The method as claimed in claim 4, wherein the step S1 further comprises: the pressure measurement signal P of the controlled pressure object is cleaned through data, error data are removed, and N times of median values are taken from continuous data to reduce fluctuation of the pressure measurement value;

step S2 further includes: calculating the pressure control deviation Pe in real time according to the pressure set value Ps and the preprocessed actual measured pressure Pr, and calculating the pressure change rate et in real time according to the unit time t and the preprocessed actual measured pressure Pr change value; according to tolerance boundary conditions set by a user: and controlling the response time TX and the fluctuation stabilization time TW, calling the optimal parameters to perform pressure precision control, and starting deep learning of the control parameters when the control response time TX and the fluctuation stabilization time TW exceed tolerance boundary conditions.

6. The method of claim 5, wherein the deep learning of parameters comprises learning of buck-boost process parameters and learning of stable process parameters; wherein:

the step of learning the buck-boost process parameters comprises the following steps: a user inputs a pressure set value, records a current pressure value P0, starts timing by the No. 1 time control, and simultaneously calculates a pressure control deviation Pe in real time; and selecting the optimal parameters from the parameter data set for pressure increase and decrease control, calculating the control response time tx when the absolute value of the pressure control deviation Pe is less than or equal to 0.005MPa, and interrupting the No. 1 time control. If the control response time TX is longer than the control response time TX set by the user, starting the parameter learning of the voltage boosting and reducing process, otherwise, continuing to control the pressure stabilizing process;

the step of learning the stability-increasing process parameters includes: performing pressure stabilization process control, and calculating the pressure change rate et in unit time in real time; and starting the timing of the No. 2 time control when the absolute value of the pressure change rate et is less than or equal to 0.002MPa, simultaneously calculating the fluctuation stable time TW, finishing the precision control of the pressure set point if the fluctuation stable time TW is more than the fluctuation stable time TW set by a user, otherwise restarting the timing of the No. 2 time control, judging, and starting the parameter learning of the stabilization process if the judgment of M times is unqualified.

7. The method of claim 6, wherein the step of learning the parameters of the buck-boost process comprises: starting control parameter learning of the voltage increasing and decreasing process, setting R to be 2, and recording a current pressure set value Ps; meanwhile, setting the pressure control set value to be P0, and calculating the pressure control deviation Pe in real time; selecting an optimal parameter from the parameter data set to perform pressure increase and decrease control;

if the absolute value of the pressure control deviation Pe is less than or equal to 0.005MPa, setting the pressure control set value as Ps, simultaneously selecting the optimal parameter from the parameter data set, and setting the pressure increasing and reducing stepping value as the current optimal parameter original stepping value R;

starting parameter learning, starting timing of a time control 3, calculating control response time tx when the absolute value of pressure control deviation Pe is less than or equal to 0.005MPa, and interrupting the time control 3;

and if the control response time TX is less than or equal to the control response time TX set by the user and the control response time TX is less than or equal to the optimal control response time txMIN, storing the current control parameter into the parameter data set and finishing the parameter learning of the voltage increasing and reducing process, otherwise, setting R to R +1 and setting the pressure control set value to P0 to perform the parameter learning of the voltage increasing and reducing process again.

8. The method of claim 6, wherein the step of learning the parameters of the stability-increasing process comprises: starting the learning of control parameters of the stabilization process, setting F to be 2, and calculating the pressure control deviation Pe and the pressure change rate et in unit time in real time;

selecting an optimal parameter from the parameter data set, and recording an original stepping value; setting an increasing and reducing pressure step value equal to the original step value of the current optimal parameter F, and performing pressure control;

if the absolute value of the pressure control deviation Pe is less than or equal to 0.005MPa and the absolute value of the pressure change rate et is less than or equal to 0.002MPa, starting the time control 4 to time, and simultaneously calculating the pressure fluctuation stabilization time tw;

and if the pressure fluctuation stabilization time TW is greater than the fluctuation stabilization time TW set by the user and the pressure fluctuation stabilization time TW is greater than the optimal fluctuation stabilization time twMAX, storing the current control parameter into the parameter data set and ending the learning of the stabilization process control parameter, otherwise, setting F +1, and continuously setting the pressure increase and decrease step value F to the current optimal parameter original step value F to perform the learning of the stabilization process control parameter again.

Technical Field

The invention belongs to the field of artificial intelligence, and particularly relates to an automatic machine learning method applied to a pressure precision control technology.

Background

The traditional automatic control is established on the basis of determining a model, and the control effect is optimal under the condition that a control object is linearly changed. The pressure control object has serious uncertainty in the control model due to factors such as rapid change of fluid medium, deformation and leakage of pipeline material, and the pressure control object is nonlinear change. When the control needs to be set at any multiple points within a wide range, the traditional automatic control of the optimal control parameters set at a certain set point can not obtain the optimal ideal control effect at other control points, even the control effect is degraded. In the industrial process, the machine learning method is applied to realize the precise control of any multipoint pressure in a wide range, and has great practical significance and theoretical value.

The invention realizes the precise control of a pressure control object by utilizing an automatic machine learning method, is innovative exploration and attempt of the automatic machine learning method, and realizes the precise control of +/-0.005 MPa deviation of any multipoint pressure control object in a range of 0-100 MPa through practical application.

Disclosure of Invention

The invention aims to realize the precise control of a pressure control object by using an automatic machine learning method in the innovative exploration and attempt of the automatic machine learning method applied to the precise control of the pressure.

The invention is realized by the following technical scheme:

an automatic machine learning system for precision pressure control, comprising: high accuracy digital pressure sensor, automatic machine learning device, gas pressure precision control device and high-pressure gas-liquid supercharging device connect gradually, wherein: the high-precision digital pressure sensor is used for measuring pressure and outputting a digital quantity signal; the automatic machine learning device comprises a data preprocessing module, a monitoring mechanism module, a data set storage module, a deep learning module and a man-machine interaction module which are connected, wherein the data preprocessing module is used for acquiring output digital quantity signals of the high-pressure gas-liquid supercharging device and removing error data; the monitoring mechanism module is used for calculating the pressure control deviation pe and the pressure change rate et in real time, calling the optimal parameters in the data set storage module and outputting digital quantity control signals according to the pressure value and the tolerance boundary condition set by a user, and automatically judging and sending a learning instruction to the deep learning module; simultaneously, optimally selecting the multiple learning results, and storing the results into a data set storage module; the data set storage module is used for storing the historical optimal parameters and the current deep learning optimal selection results and carrying out optimal selection again according to tolerance boundary conditions set by a user; the deep learning module is used for automatically learning parameters and uploading learning result data to the supervision mechanism module according to the learning instruction sent by the supervision mechanism module, if the learning result is not satisfactory, the deep learning module sends out the learning instruction again, and the deep learning module carries out secondary learning again until an end instruction is received and stops learning; the human-computer interaction module is used for providing user parameter setting and related parameter display; the gas pressure precise control device is used for receiving the digital quantity control signal and converting the digital quantity control signal into a precise control gas source signal; and constantly outputting an air source; the high-pressure gas-liquid supercharging device is used for receiving a precise control gas source signal and linearly outputting a pressure loop.

In another aspect of the present invention, an automatic machine learning method applied to pressure precision control is provided, which includes the following steps:

s1, collecting the pressure of the high-pressure gas-liquid supercharging device for measurement and outputting a digital quantity signal;

s2, calculating a pressure control deviation pe and a pressure change rate et in real time, calling optimal parameters and outputting digital quantity control signals according to a pressure value and tolerance boundary conditions set by a user, and automatically judging whether to send a deep learning instruction or not; during deep learning, optimal selection is carried out on multiple learning results at the same time; and carrying out optimal selection again according to tolerance boundary conditions set by a user;

s3, learning parameters and obtaining learning result data, if the learning result data is not satisfied, sending out the learning instruction again, learning again, and circulating N times until receiving the end instruction to stop learning, and forming a digital quantity control signal;

s4, receiving the digital control signal and converting the digital control signal into a precision control air source signal; and constantly outputting an air source;

and S5, receiving a precision control air source signal and outputting a pressure loop linearly.

The working principle and the beneficial effects of the invention are introduced as follows: through deep learning, automatic generation and historical data accumulation of optimal parameters for pressure precise control set at any multiple points in a wide range are realized, and through a supervision mechanism, the optimal control parameters are selected for pressure precise control when the pressure is controlled at any multiple points in the wide range. The method solves the key problems that the traditional automatic control has serious uncertainty in pressure control, and the nonlinear change of a control object causes difficulty in establishing an accurate mathematical model and the like. The precise control of the deviation of +/-0.005 MPa of any multipoint pressure control object in the range of 0-100 MPa is realized.

Drawings

FIG. 1 is a schematic structural view of the present invention;

FIG. 2 is a schematic diagram of the pressure fine control automatic machine learning method of the present invention;

FIG. 3 is a logic diagram of the pressure fine control of the present invention;

FIG. 4 is a logic diagram for learning control parameters for a buck-boost process in accordance with the present invention;

FIG. 5 is a logic diagram of the stable process control parameter learning of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings in conjunction with the following detailed description. It should be understood that the description is intended to be exemplary only, and is not intended to limit the scope of the present invention. Moreover, in the following description, descriptions of well-known structures and techniques are omitted so as to not unnecessarily obscure the concepts of the present invention.

Example 1: introduction to the System

An automatic machine learning system for precision pressure control, comprising: high accuracy digital pressure sensor, automatic machine learning device, gas pressure precision control device and high-pressure gas-liquid supercharging device connect gradually, wherein: the high-precision digital pressure sensor is used for measuring pressure and outputting a digital quantity signal; the automatic machine learning device comprises a data preprocessing module, a monitoring mechanism module, a data set storage module, a deep learning module and a man-machine interaction module which are connected, wherein the data preprocessing module is used for acquiring output digital quantity signals of the high-pressure gas-liquid supercharging device and removing error data; the monitoring mechanism module is used for calculating the pressure control deviation pe and the pressure change rate et in real time, calling the optimal parameters in the data set storage module and outputting digital quantity control signals according to the pressure value and the tolerance boundary condition set by a user, and automatically judging and sending a learning instruction to the deep learning module; simultaneously, optimally selecting the multiple learning results, and storing the results into a data set storage module; the data set storage module is used for storing the historical optimal parameters and the current deep learning optimal selection results and carrying out optimal selection again according to tolerance boundary conditions set by a user; the deep learning module is used for automatically learning parameters and uploading learning result data to the supervision mechanism module according to the learning instruction sent by the supervision mechanism module, if the learning result is not satisfactory, the deep learning module sends out the learning instruction again, and the deep learning module carries out secondary learning again until an end instruction is received and stops learning; the human-computer interaction module is used for providing user parameter setting and related parameter display; the gas pressure precise control device is used for receiving the digital quantity control signal and converting the digital quantity control signal into a precise control gas source signal; and constantly outputting an air source; the high-pressure gas-liquid supercharging device is used for receiving a precise control gas source signal and linearly outputting a pressure loop.

Preferably, the gas pressure precise control device comprises a digital quantity voltage signal device, a high-precision proportional pneumatic valve, a pressure stabilizing valve and a gas source; the digital quantity voltage signal device receives a digital quantity control signal through an RS485 communication port and converts the digital quantity control signal into a 0-10VDC direct current output signal, and the signal resolution is 0.0001 VDC; the electric side of the high-precision proportional pneumatic valve receives a 0-10VDC direct current signal and linearly outputs a 0-1.0MPa precision control air source signal; the pressure stabilizing valve is used for outputting the gas output by the gas source with the constant pressure of 1.0MPa and providing the gas to the side of the gas source of the high-precision proportional pneumatic valve.

Preferably, the high-pressure gas-liquid supercharging device comprises a gas-liquid supercharging pump, an output pressure loop and an oil tank; wherein the pressure ratio of the gas-liquid booster pump is 1:100, and the pressure of 0-100 MPa is linearly output to the output pressure loop.

Example 2: example demonstration

As shown in fig. 1, an automatic machine learning system for precise pressure control is composed of a high-precision digital pressure sensor, an automatic machine learning device, a gas pressure precise control device, and a high-pressure gas-liquid supercharging device which are connected in sequence;

1. the accuracy grade of the high-precision digital pressure sensor 1 is +/-0.01% FS, the measuring range is 0-100 MPa, pressure is measured, a digital quantity signal is output through an RS485 communication port, the signal resolution is 0.0001MPa, and the sampling rate is 0.1 s.

2. The automatic machine learning device 2 utilizes the high-speed processing capability of the embedded processor and the automatic machine learning technology to improve the pressure control precision. The system is formed by connecting a data preprocessing module, a monitoring mechanism module, a data set storage module, a deep learning module and a man-machine interaction module. The data preprocessing module 12 collects the pressure measurement digital quantity signals output by the high-pressure gas-liquid supercharging device 4 at a high speed, and eliminates error data through data cleaning. The number N of times of continuous data of the continuous data is obtained and can be set by a user, wherein N is not less than 3, so that the fluctuation of the pressure measurement value is reduced; the supervision mechanism module 13 calculates the pressure control deviation pe and the pressure change rate et in real time. According to the pressure value set by the user and the tolerance boundary condition, the optimal parameters in the data set storage module 14 are called and digital quantity control signals are output, and meanwhile, the learning instruction is automatically judged and sent to the deep learning module 15. Meanwhile, the optimal selection is carried out on the multiple learning results, and the results are stored in the data set storage module 14; the data set storage module 14 stores the historical optimal parameters and the current optimal selection result of deep learning, and performs optimal selection again according to tolerance boundary conditions set by the user; the deep learning module 15 automatically performs parameter learning according to the learning instruction sent by the supervision mechanism module 13 and uploads the learning result data to the supervision mechanism module 13, if the learning result is not satisfactory, the deep learning module sends out the learning instruction again, performs the second learning again, and stops learning until receiving the end instruction; the human-computer interaction module 16 mainly provides user parameter setting and related parameter display.

3. The gas pressure precision control device 3 receives a digital quantity control signal through an RS485 communication port and converts the digital quantity control signal into a 0-1.0MPa precision control gas source signal. The device consists of a digital quantity voltage signal device, a high-precision proportional pneumatic valve, a pressure stabilizing valve and an air source. The digital quantity voltage signal device 17 receives a digital quantity control signal through an RS485 communication port and converts the digital quantity control signal into a 0-10VDC direct current output signal, and the signal resolution is 0.0001 VDC; the high-precision proportional pneumatic valve 18 receives a 0-10VDC direct current signal at the electrical side and linearly outputs a 0-1.0MPa precision control air source signal; the pressure stabilizing valve 19 outputs the output gas of the gas source 20 with the constant pressure of 1.0MPa and provides the output gas to the gas source side of the high-precision proportional pneumatic valve 18.

4. The high-pressure gas-liquid supercharging device 4 receives a 0-1.0MPa precision control gas source signal and linearly outputs a 0-100 MPa pressure according to a 1:100 supercharging proportion. Consists of a gas-liquid booster pump, an output pressure loop and an oil tank. Wherein the pressurization proportion of the gas-liquid booster pump 20 is 1:100, and the pressure of 0-100 MPa is linearly output to the output pressure loop 21; the oil tank 22 provides an oil medium for the gas-liquid booster pump 20.

Example 3: introduction to the method

The automatic machine learning device is the core of the automatic machine learning method applied to pressure precise control, and selects the optimal parameters for pressure precise control through the automatic machine learning method according to the pressure control deviation Pe and the pressure change rate et. And if the pressure control response time and the pressure fluctuation stabilization time exceed the tolerance boundary conditions, starting control parameter deep learning.

An automatic machine learning method applied to pressure precise control comprises the following steps:

s1, collecting the pressure of the high-pressure gas-liquid supercharging device for measurement and outputting a digital quantity signal;

s2, calculating a pressure control deviation pe and a pressure change rate et in real time, calling optimal parameters and outputting digital quantity control signals according to a pressure value and tolerance boundary conditions set by a user, and automatically judging whether to send a deep learning instruction or not; during deep learning, optimal selection is carried out on multiple learning results at the same time; and carrying out optimal selection again according to tolerance boundary conditions set by a user;

s3, learning parameters and obtaining learning result data, if the learning result data is not satisfied, sending out the learning instruction again, learning again, and circulating N times until receiving the end instruction to stop learning, and forming a digital quantity control signal;

s4, receiving the digital control signal and converting the digital control signal into a precision control air source signal; and constantly outputting an air source;

and S5, receiving a precision control air source signal and outputting a pressure loop linearly.

Preferably, as shown in fig. 2, step S1 further includes: the pressure measurement signal P of the controlled pressure object is cleaned through data, error data are removed, and N times of median values are taken from continuous data to reduce fluctuation of the pressure measurement value;

step S2 further includes: calculating the pressure control deviation Pe in real time according to the pressure set value Ps and the preprocessed actual measured pressure Pr, and calculating the pressure change rate et in real time according to the unit time t and the preprocessed actual measured pressure Pr change value; according to tolerance boundary conditions set by a user: and controlling the response time TX and the fluctuation stabilization time TW, calling the optimal parameters to perform pressure precision control, and starting deep learning of the control parameters when the control response time TX and the fluctuation stabilization time TW exceed tolerance boundary conditions.

Preferably, the deep learning of the parameters comprises the step-up and step-down process parameter learning and the stable process parameter learning; wherein: as shown in fig. 3, a user inputs a pressure set value, records a current pressure value P0, starts timing by the time control No. 1, and calculates a pressure control deviation Pe in real time; and selecting the optimal parameters from the parameter data set for pressure increase and decrease control, calculating the control response time tx when the absolute value of the pressure control deviation Pe is less than or equal to 0.005MPa, and interrupting the No. 1 time control. If the control response time TX is longer than the control response time TX set by the user, starting the parameter learning of the voltage boosting and reducing process, otherwise, continuing to control the pressure stabilizing process;

the step of learning the stability-increasing process parameters includes: performing pressure stabilization process control, and calculating the pressure change rate et in unit time in real time; and starting the timing of the No. 2 time control when the absolute value of the pressure change rate et is less than or equal to 0.002MPa, simultaneously calculating the fluctuation stable time TW, finishing the precision control of the pressure set point if the fluctuation stable time TW is more than the fluctuation stable time TW set by a user, otherwise restarting the timing of the No. 2 time control, judging, and starting the parameter learning of the stabilization process if the judgment of M times is unqualified.

Preferably, as shown in fig. 4, starting the learning of the control parameters of the voltage increasing and decreasing process, setting R to 2, and recording the current pressure set value Ps; meanwhile, setting the pressure control set value to be P0, and calculating the pressure control deviation Pe in real time; selecting an optimal parameter from the parameter data set to perform pressure increase and decrease control;

if the absolute value of the pressure control deviation Pe is less than or equal to 0.005MPa, setting the pressure control set value as Ps, simultaneously selecting the optimal parameter from the parameter data set, and setting the pressure increasing and reducing stepping value as the current optimal parameter original stepping value R;

starting parameter learning, starting timing of a time control 3, calculating control response time tx when the absolute value of pressure control deviation Pe is less than or equal to 0.005MPa, and interrupting the time control 3;

and if the control response time TX is less than or equal to the control response time TX set by the user and the control response time TX is less than or equal to the optimal control response time txMIN, storing the current control parameter into the parameter data set and finishing the parameter learning of the voltage increasing and reducing process, otherwise, setting R to R +1 and setting the pressure control set value to P0 to perform the parameter learning of the voltage increasing and reducing process again.

Preferably, as shown in fig. 5, the stable process control parameter learning is started, F is set to 2, and the pressure control deviation Pe and the pressure change rate per unit time et are calculated in real time;

selecting an optimal parameter from the parameter data set, and recording an original stepping value; setting an increasing and reducing pressure step value equal to the original step value of the current optimal parameter F, and performing pressure control;

if the absolute value of the pressure control deviation Pe is less than or equal to 0.005MPa and the absolute value of the pressure change rate et is less than or equal to 0.002MPa, starting the time control 4 to time, and simultaneously calculating the pressure fluctuation stabilization time tw;

and if the pressure fluctuation stabilization time TW is greater than the fluctuation stabilization time TW set by the user and the pressure fluctuation stabilization time TW is greater than the optimal fluctuation stabilization time twMAX, storing the current control parameter into the parameter data set and ending the learning of the stabilization process control parameter, otherwise, setting F +1, and continuously setting the pressure increase and decrease step value F to the current optimal parameter original step value F to perform the learning of the stabilization process control parameter again.

It is to be understood that the above-described embodiments of the present invention are merely illustrative of or explaining the principles of the invention and are not to be construed as limiting the invention. Therefore, any modification, equivalent replacement, improvement and the like made without departing from the spirit and scope of the present invention should be included in the protection scope of the present invention. Further, it is intended that the appended claims cover all such variations and modifications as fall within the scope and boundaries of the appended claims or the equivalents of such scope and boundaries.