CN109366485A - A kind of field control method of online machine learning - Google Patents
A kind of field control method of online machine learning Download PDFInfo
- Publication number
- CN109366485A CN109366485A CN201811027921.1A CN201811027921A CN109366485A CN 109366485 A CN109366485 A CN 109366485A CN 201811027921 A CN201811027921 A CN 201811027921A CN 109366485 A CN109366485 A CN 109366485A
- Authority
- CN
- China
- Prior art keywords
- instruction
- controller
- data
- period
- kernel function
- 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
Links
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1628—Programme controls characterised by the control loop
- B25J9/163—Programme controls characterised by the control loop learning, adaptive, model based, rule based expert control
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1628—Programme controls characterised by the control loop
- B25J9/1653—Programme controls characterised by the control loop parameters identification, estimation, stiffness, accuracy, error analysis
Landscapes
- Engineering & Computer Science (AREA)
- Robotics (AREA)
- Mechanical Engineering (AREA)
- Mobile Radio Communication Systems (AREA)
- Feedback Control In General (AREA)
Abstract
The present invention provides a kind of field control methods of online machine learning, include the following steps: 1) model training: holding signal path is signal receiving end-field controller-instruction transmitting terminal, obtain instruction and received data that field controller is sent, it is that output valve carries out multistage fitting to insertion model using the data received as the instruction of input value, transmission, until the instruction of the data and transmission received can be fitted at least one section of kernel function, then enter next step, linear kernel function only includes the citation form of linear function；2) Developing Tactics；3) intervention control.The present invention is by way of multistage, two stages fitting, it is capable of providing effective on-line study, the investment of manpower and material resources is effectively reduced convenient for enterprise, shortens the R&D cycle, the handover of inserting controller of the smooth completion control of energy from traditional field controller to machine learning, and convenient for passing through the other long-range adjusting parameter of telecommunication mode.
Description
Technical field
The present invention relates to a kind of field control methods of online machine learning.
Background technique
Currently, application of the machine learning in Industry Control gradually increases, however local manufacturing enterprises commonly encounter most instantly
Burden is that data are seriously deficient, this causes the controller of machine learning to be difficult to train completion, and a kind of half-way house is to be divided to two
Phase carries out, and first acquires a period of time data, the code of machine learning controller is completed at the same time, then according to a small amount of number collected
According to being trained, obtain phase machine learning controller, it come into operation, continue in use acquire data it is longer when
Between, then according to the data re -training machine learning model all acquired, the second stage of machine learning controller is obtained, by the second stage of machine
Device learning controller is for finally controlling.However, in this way, a R&D cycle is too long, two carry out manpower and material resources investment
It is high, not as good as directly engagement operator for enterprise.
To solve the above problems, our company devises a kind of showing based on Time-sharing control handover control as shown in Figure 1
Station control system, the field control system can guarantee that from traditional field controller to machine learning be to control core from hardware
The controller of the heart, which carries out control handover, can be the process gradually substituted, however specifically how complete control handover,
The prior art does not provide technical inspiration.
Summary of the invention
In order to solve the above technical problems, this is online the present invention provides a kind of field control method of online machine learning
The field control method of machine learning is capable of providing effective on-line study, by way of multistage, two stages fitting convenient for enterprise
Industry is effectively reduced the investment of manpower and material resources, shortens the R&D cycle.
The present invention is achieved by the following technical programs.
A kind of field control method of online machine learning provided by the invention, includes the following steps:
1) model training: holding signal path is signal receiving end-field controller-instruction transmitting terminal, obtains scene control
The instruction and received data that device processed is sent are output valve to insertion mould using the data received as the instruction of input value, transmission
Type carries out multistage fitting, until the instruction of the data and transmission received can be fitted at least one section of kernel function, then under entering
One step, linear kernel function only include the citation form of linear function；
2) Developing Tactics: each single item adds deviant in linear kernel function, every to receive by n times field controller signal
Signal path is then switched to signal receiving end-inserting controller-instruction transmitting terminal by the hair period, is every time switched to signal path
It is one controller period of insertion when signal receiving end-inserting controller-instruction transmitting terminal, each controller cycle interpolation enters control
The output order that device processed is calculated using received data and upper controller period inserting controller according to insertion model is more
New deviant, deviant, which updates, to be completed then to enter next step, otherwise repeats the step；
3) intervention control: holding signal path is signal receiving end-inserting controller-instruction transmitting terminal, inserting controller
Insertion model is updated by being inserted into model cootrol, and in every N number of signal transmitting and receiving period.
The citation form of the insertion model Kernel Function includes at least linear function, exponential function, trigonometric function.
The deviant is initialized when being added with random value.
In the step 1), insertion model carries out multistage fitting in the following way:
A. using current demand signal period received data as input value, using the instruction that the current demand signal period sends as output
Value, traverses calculated kernel function and judges whether to be fitted, as can fitting then abandons current demand signal period received data,
As that cannot be fitted, using the instruction of received data and corresponding transmission as a data to being put into pending data library；
B. the data in pending data library are obtained to counting, such as larger than M is then by the data in pending data library to taking
Out, kernel functional parameter is calculated as available data, the elementary form of various kernel function is calculated；
C. by the data of taking-up to being substituting in the elementary form of various kernel function, at the beginning of judging whether there is any one kernel function
Etc. forms can be fitted 80% or more data pair, if any then by the elementary formal notation of the kernel function be calculated kernel function, it is right
Should kernel function fitting data to then abandoning, to pending data library is put back to, remaining kernel function reinitializes remainder data.
The standard of the fitting is that the difference between kernel function calculated result and data centering output valve is calculated less than kernel function
As a result with 10% of any one value in data centering output valve.
The N is 10~30.
The M is 10~15.
The deviant updates in the following way:
A. the instruction in current controller period is calculated according to current controller period received data；
B. the difference between the instruction in current controller period and the instruction in previous controller period is calculated；
C. according to difference retrospectively calculate deviant；
D. the instruction in current controller period is calculated according to current controller period received data again.
The deviant updates the standard completed, and calculates current controller according to current controller period received data
Gained difference, which is less than, after the instruction in period, between the instruction and the instruction in previous controller period in calculating current controller period works as
The instruction in preceding controller period and the 10% of any one of the instruction in previous controller period.
The beneficial effects of the present invention are: by way of multistage, two stages fitting, it is capable of providing effective online
It practises, the investment of manpower and material resources is effectively reduced convenient for enterprise, shortens the R&D cycle, smooth can complete control from traditional scene control
Device processed to machine learning inserting controller handover, and convenient for passing through the other long-range adjusting parameter of telecommunication mode.
Detailed description of the invention
Fig. 1 is the connection schematic diagram of field control system applied by the present invention.
Specific embodiment
Be described further below technical solution of the present invention, but claimed range be not limited to it is described.
The present invention is applied to the field control system based on Time-sharing control handover control as shown in Figure 1, and specifically one
The field control method of the online machine learning of kind, includes the following steps:
1) model training: holding signal path is signal receiving end-field controller-instruction transmitting terminal, obtains scene control
The instruction and received data that device processed is sent are output valve to insertion mould using the data received as the instruction of input value, transmission
Type carries out multistage fitting, until the instruction of the data and transmission received can be fitted at least one section of kernel function, then under entering
One step, linear kernel function only include the citation form of linear function；
2) Developing Tactics: each single item adds deviant in linear kernel function, every to receive by n times field controller signal
Signal path is then switched to signal receiving end-inserting controller-instruction transmitting terminal by the hair period, is every time switched to signal path
It is one controller period of insertion when signal receiving end-inserting controller-instruction transmitting terminal, each controller cycle interpolation enters control
The output order that device processed is calculated using received data and upper controller period inserting controller according to insertion model is more
New deviant, deviant, which updates, to be completed then to enter next step, otherwise repeats the step；
3) intervention control: holding signal path is signal receiving end-inserting controller-instruction transmitting terminal, inserting controller
Insertion model is updated by being inserted into model cootrol, and in every N number of signal transmitting and receiving period.
The citation form of the insertion model Kernel Function includes at least linear function, exponential function, trigonometric function.
The deviant is initialized when being added with random value.
In the step 1), insertion model carries out multistage fitting in the following way:
A. using current demand signal period received data as input value, using the instruction that the current demand signal period sends as output
Value, traverses calculated kernel function and judges whether to be fitted, as can fitting then abandons current demand signal period received data,
As that cannot be fitted, using the instruction of received data and corresponding transmission as a data to being put into pending data library；
B. the data in pending data library are obtained to counting, such as larger than M is then by the data in pending data library to taking
Out, kernel functional parameter is calculated as available data, the elementary form of various kernel function is calculated；
C. by the data of taking-up to being substituting in the elementary form of various kernel function, at the beginning of judging whether there is any one kernel function
Etc. forms can be fitted 80% or more data pair, if any then by the elementary formal notation of the kernel function be calculated kernel function, it is right
Should kernel function fitting data to then abandoning, to pending data library is put back to, remaining kernel function reinitializes remainder data.
The standard of the fitting is that the difference between kernel function calculated result and data centering output valve is calculated less than kernel function
As a result with 10% of any one value in data centering output valve.
The N is 10~30.
The M is 10~15.
The deviant updates in the following way:
A. the instruction in current controller period is calculated according to current controller period received data；
B. the difference between the instruction in current controller period and the instruction in previous controller period is calculated；
C. according to difference retrospectively calculate deviant；
D. the instruction in current controller period is calculated according to current controller period received data again.
The deviant updates the standard completed, and calculates current controller according to current controller period received data
Gained difference, which is less than, after the instruction in period, between the instruction and the instruction in previous controller period in calculating current controller period works as
The instruction in preceding controller period and the 10% of any one of the instruction in previous controller period.
It is usually the data of multiple sensors by the data that signal receiving end receives, in addition the control instruction received,
An input vector can be considered as, similarly, the instruction that instruction transmitting terminal is sent is typically also the controlling value of multiple executing agencies, can
To be considered as an output vector, then for input vector to the mapping relations between output vector, conventional machines can be used
The mode of habit is fitted, but due to reality in, control process may there are many state, under different conditions input vector and
Mapping relations between output vector may be different, therefore by the way of multistage fitting, can be fitted to obtain a variety of reflect
Relationship is penetrated, to meet control needs.
In mapping relations, kernel function citation form influences maximum, therefore in training study in the first stage, is inserted into mould
Type is simultaneously not involved in control, only carries out parameter update according to the data actually generated, and influence of the deviant to control is relatively small,
In the presence of traditional field controller, deviant can be corrected quickly to relatively reasonable range, and then self-adjusting is completed excellent
Change.
The present invention uses the homing method of conventional machines study, is also convenient for after the completion of modeling through other telecommunication
Mode and long-range adjusting parameter, it is more flexible for existing field controller.
Claims (9)
1. a kind of field control method of online machine learning, characterized by the following steps:
1) model training: holding signal path is signal receiving end-field controller-instruction transmitting terminal, obtains field controller
The instruction of transmission and received data, using the data that receive as the instruction of input value, transmission be output valve to insertion model into
The fitting of row multistage then enters in next step until the instruction of the data and transmission received can be fitted at least one section of kernel function
Suddenly, linear kernel function only includes the citation form of linear function；
2) Developing Tactics: each single item adds deviant in linear kernel function, every by n times field controller signal transmitting and receiving week
Signal path is then switched to signal receiving end-inserting controller-instruction transmitting terminal by the phase, and signal path is switched to signal every time
It is one controller period of insertion when receiving end-inserting controller-instruction transmitting terminal, each controller cycle interpolation enters controller
It is updated partially using received data and upper controller period inserting controller according to the output order that insertion model is calculated
Shifting value, deviant, which updates, to be completed then to enter next step, otherwise repeats the step；
3) intervention control: holding signal path is signal receiving end-inserting controller-instruction transmitting terminal, and inserting controller passes through
It is inserted into model cootrol, and updates insertion model in every N number of signal transmitting and receiving period.
2. the field control method of online machine learning as described in claim 1, it is characterised in that: the insertion model center
The citation form of function includes at least linear function, exponential function, trigonometric function.
3. the field control method of online machine learning as described in claim 1, it is characterised in that: the deviant is being added
When initialized with random value.
4. the field control method of online machine learning as described in claim 1, it is characterised in that: in the step 1), insert
Enter model and carry out multistage fitting in the following way:
A. using current demand signal period received data as input value, using the instruction that the current demand signal period sends as output valve,
It traverses calculated kernel function to judge whether to be fitted, as can fitting then abandons current demand signal period received data, such as
It cannot be fitted, using the instruction of received data and corresponding transmission as a data to being put into pending data library；
B. the data in pending data library are obtained to counting, such as larger than M makees then by the data in pending data library to taking-up
Kernel functional parameter is calculated for available data, the elementary form of various kernel function is calculated；
C. the data of taking-up are judged whether there is into the elementary shape of any one kernel function to being substituting in the elementary form of various kernel function
Formula can be fitted 80% or more data pair, if any by the elementary formal notation of the kernel function being then calculated kernel function, it is corresponding should
The data of kernel function fitting are to then abandoning, and to pending data library is put back to, remaining kernel function reinitializes remainder data.
5. the field control method of online machine learning as claimed in claim 4, it is characterised in that: the standard of the fitting
For the difference between kernel function calculated result and data centering output valve is less than in kernel function calculated result and data centering output valve
The 10% of any one value.
6. the field control method of online machine learning as described in claim 1, it is characterised in that: the N is 10~30.
7. the field control method of online machine learning as described in claim 1, it is characterised in that: the M is 10~15.
8. the field control method of online machine learning as described in claim 1, it is characterised in that: the deviant update is adopted
With such as under type:
A. the instruction in current controller period is calculated according to current controller period received data；
B. the difference between the instruction in current controller period and the instruction in previous controller period is calculated；
C. according to difference retrospectively calculate deviant；
D. the instruction in current controller period is calculated according to current controller period received data again.
9. the field control method of online machine learning as claimed in claim 1 or 8, it is characterised in that: the deviant is more
The standard newly completed is that after the instruction for calculating the current controller period according to current controller period received data, calculating is worked as
Gained difference is less than the instruction in current controller period between the instruction and the instruction in previous controller period in preceding controller period
With the 10% of any one of the instruction in previous controller period.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811027921.1A CN109366485B (en) | 2018-09-04 | 2018-09-04 | On-site control method for on-line machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811027921.1A CN109366485B (en) | 2018-09-04 | 2018-09-04 | On-site control method for on-line machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109366485A true CN109366485A (en) | 2019-02-22 |
| CN109366485B CN109366485B (en) | 2021-09-10 |
Family
ID=65404606
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811027921.1A Active CN109366485B (en) | 2018-09-04 | 2018-09-04 | On-site control method for on-line machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109366485B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100241249A1 (en) * | 2006-04-25 | 2010-09-23 | Pegasus Technologies, Inc. | System for optimizing oxygen in a boiler |
| CN106250899A (en) * | 2016-07-29 | 2016-12-21 | 华东交通大学 | A kind of orange disease and insect pests monitoring and pre-alarming method based on distributed compression perception WSN |
| CN107924384A (en) * | 2015-03-11 | 2018-04-17 | 阿雅斯迪公司 | For the system and method using study model prediction result is predicted |
| CN107991877A (en) * | 2017-12-20 | 2018-05-04 | 东南大学 | A kind of Dynamic Model Identification method and system based on Recognition with Recurrent Neural Network |
| CN108394769A (en) * | 2018-03-17 | 2018-08-14 | 北京化工大学 | elevator control method, server and computer readable storage medium |
-
2018
- 2018-09-04 CN CN201811027921.1A patent/CN109366485B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100241249A1 (en) * | 2006-04-25 | 2010-09-23 | Pegasus Technologies, Inc. | System for optimizing oxygen in a boiler |
| CN107924384A (en) * | 2015-03-11 | 2018-04-17 | 阿雅斯迪公司 | For the system and method using study model prediction result is predicted |
| CN106250899A (en) * | 2016-07-29 | 2016-12-21 | 华东交通大学 | A kind of orange disease and insect pests monitoring and pre-alarming method based on distributed compression perception WSN |
| CN107991877A (en) * | 2017-12-20 | 2018-05-04 | 东南大学 | A kind of Dynamic Model Identification method and system based on Recognition with Recurrent Neural Network |
| CN108394769A (en) * | 2018-03-17 | 2018-08-14 | 北京化工大学 | elevator control method, server and computer readable storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 陈海列等: "一种基于DCS的分时训练神经网络", 《自动化应用》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109366485B (en) | 2021-09-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Griffith et al. | Policy shaping: Integrating human feedback with reinforcement learning | |
| CN108121216B (en) | Automated workshop virtual debugging method based on digital factory | |
| CN107303675A (en) | Optimize the cell controller of the action of the production system with many industrial machineries | |
| EP3385796B1 (en) | Control device | |
| Steiger et al. | An improved batch means procedure for simulation output analysis | |
| CN107085373A (en) | A kind of cross-coupling control method of the networking multi-shaft motion control system based on Communication Jamming observer | |
| WO2005019949A1 (en) | Pid parameter adjustment device | |
| CN111222553B (en) | Training data processing method and device of machine learning model and computer equipment | |
| CN115660115A (en) | Method, device and equipment for training federated learning model and storage medium | |
| US7529651B2 (en) | Accurate linear parameter estimation with noisy inputs | |
| CN109366485A (en) | A kind of field control method of online machine learning | |
| CN105637810B (en) | Distance-measuring device and distance measurement method | |
| CN116560322B (en) | Production control method and system for assembly production line | |
| CN106385376B (en) | A kind of method for controlling network congestion based on Continuum Model | |
| CN108563565B (en) | Method for establishing reliability quantitative analysis model of flight landing guidance system | |
| CN111555793A (en) | Intelligent autonomous timing method and system for satellite wireless communication network | |
| CN109188904A (en) | A kind of on-line training field control method based on machine learning | |
| CN111045938A (en) | Reliability modeling method for introducing open-source software based on Pareto distributed faults | |
| CN109669348B (en) | Industrial robot based on iterative feedback adjustment and trajectory tracking control method thereof | |
| CN109240227A (en) | A kind of field control method based on Time-sharing control handover control | |
| CN101770532A (en) | Method for improving circuit emulator convergence | |
| CN109325593B (en) | Method for on-site control time-sharing control handover control right | |
| US7587710B2 (en) | Method for determining the processing sequence of function blocks of an automated system and corresponding automated system | |
| Ting et al. | Analysis and design of the time-delay compensation for networked control systems with random communication delay | |
| SG10201808290XA (en) | Systems and methods for optimization of data center cooling |
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 | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20211130 Address after: 535300 the Guangxi Zhuang Autonomous Region Qinzhou Pubei County Town Industrial Zone Patentee after: GUANGXI KAIXING CREATIVE TECHNOLOGY CO.,LTD. Address before: 530200 No. 8, Ting Ting Road, Yongning District, Nanning, the Guangxi Zhuang Autonomous Region Patentee before: NANNING University |