CN111624954A - Advanced process control machine learning framework for flow type manufacturing industry - Google Patents

Advanced process control machine learning framework for flow type manufacturing industry Download PDF

Info

Publication number
CN111624954A
CN111624954A CN201910185629.0A CN201910185629A CN111624954A CN 111624954 A CN111624954 A CN 111624954A CN 201910185629 A CN201910185629 A CN 201910185629A CN 111624954 A CN111624954 A CN 111624954A
Authority
CN
China
Prior art keywords
data
control
advanced
machine learning
industry
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910185629.0A
Other languages
Chinese (zh)
Other versions
CN111624954B (en
Inventor
孔泉
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Miaowei Hangzhou Technology Co ltd
Original Assignee
Miaowei Hangzhou Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Miaowei Hangzhou Technology Co ltd filed Critical Miaowei Hangzhou Technology Co ltd
Priority to CN201910185629.0A priority Critical patent/CN111624954B/en
Priority claimed from CN201910185629.0A external-priority patent/CN111624954B/en
Publication of CN111624954A publication Critical patent/CN111624954A/en
Application granted granted Critical
Publication of CN111624954B publication Critical patent/CN111624954B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G05CONTROLLING; REGULATING
    • G05BCONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B19/00Programme-control systems
    • G05B19/02Programme-control systems electric
    • G05B19/418Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS], computer integrated manufacturing [CIM]
    • G05B19/41885Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS], computer integrated manufacturing [CIM] characterised by modeling, simulation of the manufacturing system

Abstract

The invention discloses a machine learning framework for advanced process control of flow type manufacturing industry, which comprises a model library, a data reading and preprocessing module, an offline training engine, a real-time model performance evaluation component and an advanced process controller, wherein the model library is used for storing data; the invention can automatically realize the functions of feature extraction, data modeling, online control, real-time monitoring and parameter updating and the like, and in addition, the framework can help a user to find the most necessary data item for high-level process control and help a factory to select a proper instrument.

Description

Advanced process control machine learning framework for flow type manufacturing industry
Technical Field
The invention relates to the field of intelligent control in process control, in particular to a machine learning framework for advanced process control of flow type manufacturing industry.
Background
By applying intelligent control in the process type manufacturing industry, the intelligent manufacturing level can be greatly improved, the production efficiency is improved, and the production cost is reduced. Existing Advanced Process Control (APC) systems often rely on an ideal equation of state model of the controlled object or require a strong industry expertise background for object modeling to perform predictive control. The method has high requirements on the knowledge reserve of the industry and needs strong actual scene strain capacity. In addition, in practical deployment applications, the existing APC solution also requires an engineer to adjust and optimize parameters on site according to actual scenes, and these relevant parameters also need to be continuously adjusted to ensure that the expression capability of the model does not degrade as time goes on and the machine equipment ages.
On the other hand, the intellectualization of the whole process type manufacturing industry is in a development stage at present, and the situations that the whole informatization level is not high, a large number of online instruments are lost or the performance is poor can occur; and the manufacturing enterprise does not know how to select and configure the necessary meters for informatization/intelligentization modification. How to quickly evaluate whether the on-line measuring instrument is necessary, reasonable and accurate is another problem influencing the intelligent modification of the process type manufacturing industry.
Disclosure of Invention
To address the above-identified problems, the present invention provides a machine learning framework for advanced process control for flow-type manufacturing.
In order to achieve the purpose, the invention adopts the following technical scheme:
a machine learning framework for advanced process control of flow type manufacturing industry comprises a model library, a data reading and preprocessing module, an offline training engine, a real-time model performance evaluation component and an advanced process controller;
the model library is internally provided with a plurality of machine learning models suitable for the process industry,
the data reading and preprocessing module is responsible for acquiring data and preprocessing the data;
the off-line training engine selects one or more models in the model base and the preprocessed data to search the optimal model and train the model parameters;
the real-time evaluation component of the model performance can acquire field operation data in real time, test the model after training is completed, and select a model with optimal performance in actual test to perform actual process control;
and the advanced process controller finds the optimal solution of the control variable by using the screened optimal model and the field real operation data, and outputs the optimal solution to a hardware control unit of the equipment.
Further, machine learning models include, but are not limited to, standard linear regression, polynomial regression, Lasso regression, locally weighted regression, random forest, gradient rise decision trees.
Further, the data reading-in and preprocessing module reads data from historical data.
Further, the data reading and preprocessing module acquires data in an automatic excitation mode.
Further, the preprocessing method of the data reading-in and preprocessing module comprises the steps of filtering illegal data and generating new data at a high latitude.
Further, the offline training engine may also help the user find the multiple data items that are most relevant to the final control objective.
Further, the real-time assessment component of model performance is responsible for switching to a more optimal model or issuing a system exception notification when a decrease in performance of the model is found or begins to decrease during actual operation.
The invention has the beneficial effects that:
(1) the invention provides a machine learning framework for advanced process control of flow type manufacturing industry, which does not need the basic background professional knowledge of the industry, can automatically realize the functions of feature extraction, data modeling, online control, real-time monitoring and parameter updating and the like, and can help a user to find out the most necessary data items for advanced process control, such as whether online measuring instruments are needed or not, or find out which data items with abnormal fluctuation, such as the possible problems of the performance of the online instruments, thereby helping a factory to select proper instruments and saving the hardware cost. The framework is designed for flow-type manufacturing and has passed typical pilot verification.
(2) The invention greatly accelerates the development speed of high-level process control in the process type manufacturing industry; because the modeling is established on the basis of data driving, high-precision modeling can be carried out without professional background knowledge; the online maintenance of the model can be automatically carried out, and the long-time stable operation is ensured without human intervention; the method can help select a reasonable online instrument scheme in the intelligent modification of a factory, thereby reducing the cost.
Drawings
FIG. 1 is a schematic diagram of a learning framework of the present invention.
Detailed Description
The technical solutions of the present invention are further described in detail below with reference to the accompanying drawings, and it should be noted that the detailed description is only for describing the present invention, and should not be construed as limiting the present invention.
The invention relates to a machine learning framework for advanced process control of flow type manufacturing industry, which comprises a model library, a data reading and preprocessing module, an offline training engine, a real-time model performance evaluation component and an advanced process controller, wherein the model library is shown in figure 1;
the model base is internally provided with a plurality of machine learning models suitable for the process industry, including but not limited to standard linear regression, polynomial regression, LASSO regression, local weighted regression, random forest and gradient ascending decision trees;
the data reading and preprocessing module can read data from historical data (the historical data comprises a database, a file and the like, and the historical data is all stored data) or acquire the data in an automatic excitation mode; after the data are obtained, the module can also preprocess the data, filter illegal data and generate new data with high latitude; automatic excitation: the control signal or the combination of the control signals is provided experimentally, and the excitation generation of the control signals is realized by a permutation and combination algorithm plus conditional constraint, so that the safety and the high efficiency of the whole process are ensured;
the off-line training engine selects one or more models in the model base and the preprocessed data to search the optimal model and train the model parameters; all training processes can realize automatic operation by simple configuration, wherein the content of the simple configuration comprises the target precision expected to be achieved or the limit of the number of calculation iterations, the allowable variable range of model parameters, which optimization mode is used and the like; the engine can also help a user to discover a plurality of data items with the maximum correlation with the final control target, or discover which data items cause instability of the model, so that the most appropriate online instrument is helped to be selected, and hardware transformation cost is saved;
the real-time model performance evaluation component can acquire real-time on-site running data, test the model after training is completed, and select a model with optimal performance in actual test to perform actual process control; when the performance of the model is found to be reduced or begins to be reduced in the actual operation process, the component is responsible for switching a better model or sending out a system exception notice;
and inputting the control target into the advanced process controller, automatically calculating and finding the optimal solution of the control variable by the advanced process controller by using the screened optimal model and the real field operation data, and outputting the optimal solution to a hardware control unit (such as PID control input) of the equipment.
Example 1
The invention is applied to the evaporation process of a certain large nonferrous metal manufacturing plant, and finally realizes a solution of advanced process control through the framework, the control precision and the stability are improved by more than 10 times compared with the previous solution, and the long-term stable work exceeds 1 year.
The concrete implementation components are as follows:
1) in this embodiment, the data set is historical data of various original devices, such as historical data of various online meters, and the historical data is data displayed by the various online meters;
2) the data reading and preprocessing module performs differential preprocessing on the data in the step (1) to obtain a new data set;
3) performing feature extraction on the data set in the step (2) through a LASSO model, wherein the extracted features are as follows: finding a modeling data item by using the mother liquor temperature, the mother liquor flow, the temperature of a five-effect heating chamber, the temperature of a six-effect heating chamber, the temperature of a one-effect discharge material, the temperature of a five-effect discharge material, the liquid level of a three-flash evaporator, the liquid level of a four-flash evaporator, the liquid level of a one-effect evaporator and the liquid level of a five-effect evaporator, wherein the extracted characteristic is the modeling data item;
the characteristic extraction method has various methods, such as random forest, GBDT, Pearson coefficient and the like, the LASSO model is selected for characteristic extraction in the embodiment, the LASSO model can achieve the required effect, and the method is simple and rapid;
in other embodiments, other preprocessing modes and other feature extraction methods may be employed;
in other embodiments, it may also be: firstly, extracting the characteristics of historical data, then carrying out differential preprocessing on the extracted data to obtain a new data set, and then extracting the characteristics of the new data set through an LASSO model;
4) the off-line training engine selects one or more models in the model base and modeling data items to search the optimal model and train model parameters;
in this embodiment, modeling is performed by standard linear regression;
according to different parameters (the parameters can be temperature, humidity, concentration and the like), different models can be established and serve as standby models, and in the embodiment, the standby models comprise a gradient ascending decision tree and a local weighted regression besides standard linear regression;
5) the real-time evaluation component of the model performance obtains real-time on-site operation data, tests the model after the training in the step 4), and selects a model with optimal performance in actual test to perform actual process control; when the performance of the model is found to be reduced or begins to be reduced in the actual operation process, the component is responsible for switching a better model or sending out a system exception notice; in this embodiment, the component searches for a better model from the standby model;
6) and inputting a control target (the control target can be the solution concentration controlled within a certain range) into the advanced process controller, automatically calculating and finding the optimal solution of the control variable by the advanced process controller by using the screened optimal model and the real field operation data, and outputting the optimal solution to a hardware control unit of the equipment to realize the control purpose.
Example 2
The invention is applied to the raw material process of a certain large-scale nonferrous metal manufacturing plant, and helps to find the performance defects of the online instrument through the frame and the data analysis, thereby avoiding unnecessary hardware purchase and reducing the cost;
the concrete implementation components are as follows:
1) the data set is historical data from a laboratory and real-time meter data;
2) a pretreatment operation and a feature extraction operation step similar to those of embodiment 1;
3) the correlation coefficient of the instrument data is very low through both the LASSO model and the Pearson coefficient; for example, it is not normal to find that the correlation coefficient of the meter data for measuring temperature and the meter data for measuring concentration is low, and generally, the correlation coefficient of temperature and concentration should be high, which may be caused by a problem in the meter, a broken pipeline or other situations;
4) modeling the laboratory historical data operated in the step (2), modeling the real-time instrument data operated in the step (2), and verifying to obtain: the quality of the instrument data is poor;
5) by model parameter tracing back, it was confirmed that: the difference between the meter data and the actual situation is large and unstable; this illustrates a problem with the meter.

Claims (7)

1. A machine learning framework for advanced process control of flow type manufacturing industry is characterized by comprising a model library, a data reading and preprocessing module, an offline training engine, a real-time model performance evaluation component and an advanced process controller;
a plurality of machine learning models suitable for the process industry are built in the model bank;
the data reading and preprocessing module is responsible for acquiring data and preprocessing the data;
the off-line training engine selects one or more models in the model base and the preprocessed data to search the optimal model and train the model parameters;
the real-time evaluation component of the model performance can acquire field operation data in real time, test the model after training is completed, and select a model with optimal performance in actual test to perform actual process control;
and the advanced process controller finds the optimal solution of the control variable by using the screened optimal model and the field real operation data, and outputs the optimal solution to a hardware control unit of the equipment.
2. The machine learning framework of advanced process control for flow-based manufacturing industry of claim 1, wherein the machine learning models include but are not limited to standard linear regression, polynomial regression, Lasso regression, locally weighted regression, random forest, gradient rising decision tree.
3. The machine learning framework of advanced process control for flow-type manufacturing industry of claim 1, wherein the data reading and preprocessing module reads data from historical data.
4. The machine learning framework of advanced process control for flow-type manufacturing industry as claimed in claim 1, wherein the data reading and preprocessing module obtains data by way of automatic activation.
5. The machine learning framework of advanced process control for flow-type manufacturing industry as claimed in claim 1, wherein the preprocessing method of the data reading and preprocessing module includes filtering illegal data and generating new data at high latitude.
6. The machine learning framework of advanced process control for flow-type manufacturing industry as claimed in claim 1, wherein the offline training engine can also help the user to find the data items with the greatest correlation with the final control objective.
7. The machine learning framework of advanced process control for flow-type manufacturing as claimed in claim 1, wherein the real-time model performance evaluation component is responsible for switching better models or issuing system exception notifications when model performance degradation or start degradation is found during actual operation.
CN201910185629.0A 2019-03-12 Advanced process control machine learning system for flow type manufacturing industry Active CN111624954B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910185629.0A CN111624954B (en) 2019-03-12 Advanced process control machine learning system for flow type manufacturing industry

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910185629.0A CN111624954B (en) 2019-03-12 Advanced process control machine learning system for flow type manufacturing industry

Publications (2)

Publication Number Publication Date
CN111624954A true CN111624954A (en) 2020-09-04
CN111624954B CN111624954B (en) 2021-06-04

Family

ID=

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102004444A (en) * 2010-11-23 2011-04-06 华东交通大学 Multi-model predictive control method for component content in process of extracting rare earth
CN102998973A (en) * 2012-11-28 2013-03-27 上海交通大学 Multi-model self-adaptive controller of nonlinear system and control method
CN106094517A (en) * 2016-06-17 2016-11-09 上海环保工程成套有限公司 A kind of Multi model Predictive Controllers of variable working condition sewage disposal process
CN106249724A (en) * 2016-09-14 2016-12-21 东北大学 A kind of blast furnace polynary molten steel quality forecast Control Algorithm and system
CN108241348A (en) * 2018-01-09 2018-07-03 北京科技大学 A kind of industrial process of data-driven monitors in real time and fault detection method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102004444A (en) * 2010-11-23 2011-04-06 华东交通大学 Multi-model predictive control method for component content in process of extracting rare earth
CN102998973A (en) * 2012-11-28 2013-03-27 上海交通大学 Multi-model self-adaptive controller of nonlinear system and control method
CN106094517A (en) * 2016-06-17 2016-11-09 上海环保工程成套有限公司 A kind of Multi model Predictive Controllers of variable working condition sewage disposal process
CN106249724A (en) * 2016-09-14 2016-12-21 东北大学 A kind of blast furnace polynary molten steel quality forecast Control Algorithm and system
CN108241348A (en) * 2018-01-09 2018-07-03 北京科技大学 A kind of industrial process of data-driven monitors in real time and fault detection method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
翟军勇: "基于动态模型库的多模型切换控制", 《控制理论与应用》 *

Similar Documents

Publication Publication Date Title
US20190286076A1 (en) Process Model Identification in a Process Control System
JP5833088B2 (en) Process controller and advanced control element generation system
JP5933485B2 (en) Robust process model identification method and system in model-based control techniques
US8046096B2 (en) Analytical server integrated in a process control network
CN101925866B (en) There is the adaptive model predictive controller of the robust of adjustment for compensation model mismatch
US7949417B2 (en) Model predictive controller solution analysis process
US7194317B2 (en) Fast plant test for model-based control
CN101908495B (en) System and method for implementing a virtual metrology advanced process control platform
US8036760B2 (en) Method and apparatus for intelligent control and monitoring in a process control system
US10809153B2 (en) Detecting apparatus, detection method, and program
JP5111719B2 (en) Method and system for collecting and retrieving time-series real-time and non-real-time data
US9164501B2 (en) Methods and apparatus to manage data uploading in a process control environment
JP5307492B2 (en) Process model history management method and selection method
US20170371311A1 (en) Using soft-sensors in a programmable logic controller
US20070078530A1 (en) Method and system for controlling a batch process
KR20130071369A (en) Method for screening samples for building prediction model and computer program product thereof
EP2118711B1 (en) Apparatus and method for automated closed-loop identification of an industrial process in a process control system.
TWI385492B (en) A system for maintaining and analyzing manufacturing equipment and method therefor
KR101998553B1 (en) Prediction Method of Short-Term Wind Speed and Wind Power and Power Supply Line Voltage Prediction Method Therefore
US20160171037A1 (en) Model change boundary on time series data
US7890200B2 (en) Process-related systems and methods
WO2007024847A2 (en) Adaptive multivariable mpc controller
EP3112961B1 (en) Control parameter optimizing system and operation control optimizing apparatus equipped therewith
US20070225835A1 (en) Computer method and apparatus for adaptive model predictive control
DE112011101738T5 (en) Multi-level process modeling process

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant