CN106503798A - Based on rough set and the method for diagnosing faults of the pump of BP neural network - Google Patents
Based on rough set and the method for diagnosing faults of the pump of BP neural network Download PDFInfo
- Publication number
- CN106503798A CN106503798A CN201510566043.0A CN201510566043A CN106503798A CN 106503798 A CN106503798 A CN 106503798A CN 201510566043 A CN201510566043 A CN 201510566043A CN 106503798 A CN106503798 A CN 106503798A
- Authority
- CN
- China
- Prior art keywords
- neural network
- pump
- training sample
- yojan
- decision table
- 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.)
- Pending
Links
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The present invention relates to a kind of method for diagnosing faults based on rough set and the pump of BP neural network, comprises the following steps：The training sample characteristic of benchmark is collected, the decision table after corresponding yojan is obtained, is set up BP neural network model；By training sample characteristic to be diagnosed, the decision table after corresponding yojan is obtained, be input to the BP neural network model, carry out fault diagnosis.The present invention by being combined BP neural network with coarse central algorithm, first with the dimension that rough set theory effectively reduces pump dynagraoph sample characteristics, recycle sample characteristics construction BP neural network evaluator after yojan, simplify BP neural network structure, the time of the study of BP neural network evaluator and operation is reduced, nicety of grading is improved.
Description
Technical field
The present invention relates to petrochemical industry and wireless senser field, specifically a kind of based on rough set with
The method for diagnosing faults of the pump of BP neural network.
Background technology
Oil as a kind of non-renewable resources, for country political, economical, military have irreplaceable
Strategic importance.As China's economy is lasting, stable, quickly development, oil consumption also sustainable growth,
Crude oil demand persistently rises, and oil insufficiency of supply-demand has the trend of increase, and is increasingly becoming restriction China economy
One of key factor of exhibition.Ended for the end of the year 2009, residual recoverable reserves is verified for 27.9 hundred million tons by CNPC,
Residual recoverable reserves reserve-production ratio is 14.8, but in these residual recoverable reserves, petroleum resources lay in quality
Poor, hypotonic, special hypotonic or ultralow permeability, viscous crude and buried depth are more than the petroleum resources of 3500m more than 50%,
Not only exploitation difficulties in exploration is being gradually increased, and development cost increases and original oil zone comprehensive water cut is high, generally enters
Enter the production decline stage.Such severe situation is faced, each oilfield enterprise will appreciate that raising crude oil production
The importance of efficiency, and do the aspects such as investment, cost-effective, raising oil field digital management level are reduced
Substantial amounts of effort.Wherein, it is an important embodiment WIA technology to be applied to oil field production figuresization management.
Working condition of the working condition energy reaction pump of pump in deep under ground, the diagnosis of pump dynagraoph is to analyzing pump
Working condition is most important, and when underground equipment breaks down, pump dynagraoph can show a certain special shape.
Therefore, the set feature of pump dynagraoph is the Main Basiss for carrying out Fault Identification, by calculating, analyzing, with regard to energy
Determine the failure of underground equipment.
Application of the neutral net in fault diagnosis is more and more extensive, has become intelligent trouble diagnosis field
Study hotspot, but the limitation that there is also based on the diagnostic method of artificial neural network, such as：Need more
Different classes of training example is only possible to for neural network learning so that network bangle, relatively steady so as to draw
Fixed result；For the complicated system that is diagnosed, the amount of calculation needed for training more than each node layer number of network, can be made
More with the time, it is impossible to which that diagnostic result is made explanations；Artificial neural network structure, parameter setting, training
The multifactor precision and generalization ability to artificial neural network such as size, sample quality has a direct impact.
Rough set theory is a kind of for processing the mathematical method for not completing inexact knowledge, and the theory need not
With regard to any initial or additional information of data, process is analyzed to imperfect imprecise data directly,
In fault diagnosis system, rough set theory simplifies the conditional attribute and decision-making category for obtaining by simple decision table
Dependence between property, and by removing redundant attributes, knowledge representation space dimensionality can be greatly simplified.
Thus it is highly significant that neutral net is combined with rough set theory, and do not have a kind of method now by two
Person is effectively combined together.
Content of the invention
For the deficiencies in the prior art, the present invention provides a kind of ground gathered based on WIA-PA wireless indicators
Indicator card, is combined with coarse central algorithm using BP neural network, carries out pre- place using rough set theory to data
Reason, extracts key element as the input of network, so as to simplify BP neural network structure, improves nicety of grading
A kind of method for diagnosing faults.
The technical scheme that adopted for achieving the above object of the present invention is：
A kind of method for diagnosing faults based on rough set and the pump of BP neural network, comprises the following steps：Collect
The training sample characteristic of benchmark, obtains the decision table after corresponding yojan, sets up BP neural network model；
By training sample characteristic to be diagnosed, the decision table after corresponding yojan is obtained, be input to the BP nerves
Network model, carries out fault diagnosis.
The decision table obtained after corresponding yojan is comprised the following steps：
Step 1：Collect the training sample pump dynagraoph of benchmark and binary conversion treatment is carried out to which, obtain training sample
Pump dynagraoph curve；
Step 2：Corresponding grid search-engine vector is formed according to training sample pump dynagraoph curve；
Step 3：By being compared to the grid search-engine vector of different training sample pump dynagraophs, different works are drawn
The conditional attribute and decision attribute of condition condition, sets up decision table；
Step 4：Yojan is carried out to decision table, the decision table after yojan is obtained.
Described form corresponding grid search-engine vector process be：Training sample pump dynagraoph curve is divided into
M × n grid, wherein m=2n, the grid that training sample pump dynagraoph curved boundary is passed through is entered as
" 1 ", the grid not passed through are entered as " 0 ", and all for border inner meshes are all entered as " 1 ", obtain
Grid search-engine vector.
Described yojan is carried out to decision table include procedure below：
Step 1：Calculate the positive domain POS of C of DC(D) number comprising training sample in, wherein C are condition
Attribute, D are decision attribute；
Step 2：It is empty set to make R=φ, φ, to C each attribute P in R, calculate POS respectivelyC∪{p}(D)
In the training sample number that includes；
Step 3：Selection makes POSR∪{p}(D) comprising the attribute that training sample number is most in, remember P',
R=R ∪ { P'}；
Step 4：If POSR∪{p}(D) the training sample number included in is equal to POSC(D) training included in
Number of samples, then export minimum fault diagnosis character subset, the as decision table after yojan, otherwise return step
Rapid 1；
The BP neural network model of setting up includes procedure below：
Step 1：Weights and the initialization of threshold values, by initial weight and threshold values between each for BP neural network layer
Random is assigned to [0,1] interval value；
Step 2：By the yojan decision table of the training sample of benchmark, as the input vector of BP neural network；
Step 3：Rule of thumb formula determines BP neural network structure：
Wherein, P is concealed nodes number, and r is input layer number, and s is output layer nodes；
Step 4：Weight matrix between implicit node and BP neural network are calculated by Sigmoid functions
Output, completes the foundation of BP neural network model.
The training sample includes following operating mode：Normal work, gases affect, feed flow are not enough, sucker rod breaks
De-, oil is thick, travelling valve leakage, touch on pump, touch under pump, standing valve leakage, plunger abjection seating nipple,
Gas lock, the leakage of double valves, holddown, sand production, wax deposition.
The invention has the advantages that and advantage：
The present invention is effectively dropped first with rough set theory by combining BP neural network with coarse central algorithm
The dimension of low pump work pattern eigen, recycles sample characteristics construction BP neural network evaluator after yojan, letter
Change BP neural network structure, reduce the time of the study of BP neural network evaluator and operation, improve nicety of grading.
Description of the drawings
Fig. 1 is method of the present invention flow chart；
Fig. 2 is the ground work(figure and pump dynagraoph of the feed flow deficiency of the present invention；
Specific embodiment
Below in conjunction with the accompanying drawings and embodiment the present invention is described in further detail.
It is illustrated in figure 1 method of the present invention flow chart.
Based on rough set and the oil well fault diagnostic method of neutral net, surface dynamometer card is gathered, is being obtained
After its pump dynagraoph, fault diagnosis is carried out to pump dynagraoph using rough set and BP neural network.
Step 1：15 kinds of training samples are collected, including：Normal work, gases affect, feed flow are not enough, take out
Beam hanger is disconnected de-, and oil is thick, and travelling valve is missed, and touches, touch under pump on pump, and standing valve is missed, and " plunger is deviate from
Seating nipple, gas lock, double valve leakages, holddown, sand production, wax deposition.
Step 2：Binary conversion treatment is carried out to pump dynagraoph, the image values matrix after process is made up of 0 and 1.
Step 3：The maximum matrix area that Pixel of Digital Image is 0 is intercepted, by the image in this region through collection
Conversion is closed, the bianry image for becoming 32*32 is allowed to, bianry image is carried out inverse process then, with so
The vector that the numeral 0,1 of each pixel of the image that obtains is constituted.
Step 4：On curve along curve the multiple characteristic points of formation direction acquisition order, thus obtained song
Coordinate vector X, Y of characteristic point on line, then interpolation on another abscissa, it is possible to draw enough
Point coordinates, connects these points and just delineates pump dynagraoph curve.
Step 5：Given sample standard deviation through load and the binary image of unique normalized, pump work
Figure is divided into m × n (wherein, m=2n) individual grid, and the grid that pump dynagraoph curved boundary is passed through assigns " 1 ",
The grid not passed through assigns " 0 ", and all for border inner meshes are all entered as " 1 ", therefore by all grids
Eigenvalue cluster is combined the grid search-engine vector to form m × n dimension, and grid search-engine can react pump dynagraoph
Global shape distribution.
The binaryzation assignment matrix of 1 pump dynagraoph of table
Step 6：Set up decision table and attribute discretization：
The decision table of 2 sample set of table
Step 7：Attribute reduction：
(1) in decision table T, conditional attribute is C, and decision attribute is D, calculates the positive domain POS of C of DC(D)
In comprising sample number.
(2) R=φ are made, to conditional attribute C R repeat：To C each attribute P in R, calculate
POSC∪{p}(D) number of samples included in；Selection makes POSR∪{p}(D) most comprising number of samples in
Attribute, remembers P', R=R ∪ { p }；New R values are assigned to former R, also new R values can be represented
For R '；If POSR∪{p}(D) number of samples included in is equal to POSC(D) number of samples included in, then
The minimum fault diagnosis character subset of output；Recalculate (1) for otherwise returning to step 7.
Step 8：The design of BP neural network model：The corresponding instruction of the minimal condition property set that obtained with yojan
Practice sample to learn BP neural network and trained, by rough set and the BP neural network of network parallel
Minimal decision-making regulation is realized in study, and the correction of decision scheme is by the friendship between BP neural network and rough set study
Change improvement, until rough set study select the decision ruless that minimum attribute constitutes all correctly can divide all of
Till test set sample.
(1) initialization of weights and threshold values, will be random for the initial weight and threshold values between each for BP networks layer
It is assigned to [0,1] interval value.
(2) learning sample, input vector of the sample that rough set is screened as BP neural network are input into.
(3) determine BP neural network structure：Rule of thumb formula
P be concealed nodes number, r and s be respectively input layer, output layer nodes.
(4) BP neural network function is selected, is calculated between implicit node by the Sigmoid functions that selects
Weight matrix and the output of BP neural network.
Step 9：Fault diagnosis, to the new data being input into, makees same pretreatment, i.e., only takes minimal condition
The attribute of property set, makees identical normalization.Input is trained to sample to the BP neural network for training, defeated
Go out to recognize classification results.
Claims (6)
1. a kind of method for diagnosing faults based on rough set and the pump of BP neural network, it is characterised in that include with
Lower step：The training sample characteristic of benchmark is collected, the decision table after corresponding yojan is obtained, BP god is set up
Through network model；By training sample characteristic to be diagnosed, the decision table after corresponding yojan is obtained, be input into
To the BP neural network model, fault diagnosis is carried out.
2. the method for diagnosing faults based on rough set and the pump of BP neural network according to claim 1, its
It is characterised by, the decision table obtained after corresponding yojan is comprised the following steps：
Step 1：Collect the training sample pump dynagraoph of benchmark and binary conversion treatment is carried out to which, obtain training sample
Pump dynagraoph curve；
Step 2：Corresponding grid search-engine vector is formed according to training sample pump dynagraoph curve；
Step 3：By being compared to the grid search-engine vector of different training sample pump dynagraophs, different works are drawn
The conditional attribute and decision attribute of condition condition, sets up decision table；
Step 4：Yojan is carried out to decision table, the decision table after yojan is obtained.
3. the method for diagnosing faults based on rough set and the pump of BP neural network according to claim 2, its
It is characterised by：Described form corresponding grid search-engine vector process be：By training sample pump dynagraoph curve point
Into m × n grid, wherein m=2n, the grid assignment that training sample pump dynagraoph curved boundary is passed through
For " 1 ", the grid not passed through is entered as " 0 ", and all for border inner meshes are all entered as " 1 ", obtains
Grid search-engine vector is arrived.
4. the method for diagnosing faults based on rough set and the pump of BP neural network according to claim 2, its
It is characterised by：Described yojan is carried out to decision table include procedure below：
Step 1：Calculate the positive domain POS of C of DC(D) number comprising training sample in, wherein C are condition
Attribute, D are decision attribute；
Step 2：It is empty set to make R=φ, φ, to C each attribute P in R, calculate POS respectivelyC∪{p}(D)
In the training sample number that includes；
Step 3：Selection makes POSR∪{p}(D) comprising the attribute that training sample number is most in, remember P',
R=R ∪ { P'}；
Step 4：If POSR∪{p}(D) the training sample number included in is equal to POSC(D) training included in
Number of samples, then export minimum fault diagnosis character subset, the as decision table after yojan, otherwise return step
Rapid 1.
5. the method for diagnosing faults based on rough set and the pump of BP neural network according to claim 1, its
It is characterised by：The BP neural network model of setting up includes procedure below：
Step 1：Weights and the initialization of threshold values, by initial weight and threshold values between each for BP neural network layer
Random is assigned to [0,1] interval value；
Step 2：By the yojan decision table of the training sample of benchmark, as the input vector of BP neural network；
Step 3：Rule of thumb formula determines BP neural network structure：
Wherein, P is concealed nodes number, and r is input layer number, and s is output layer nodes；
Step 4：Weight matrix between implicit node and BP neural network are calculated by Sigmoid functions
Output, completes the foundation of BP neural network model.
6. the fault diagnosis based on rough set and the pump of BP neural network according to any one of Claims 1 to 5
Method, it is characterised in that：The training sample includes following operating mode：Normal work, gases affect, feed flow
Deficiency, rod parting, oil thick, travelling valve leakage, touch on pump, touch under pump, standing valve leakage,
Plunger abjection seating nipple, gas lock, the leakage of double valves, holddown, sand production, wax deposition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510566043.0A CN106503798A (en) | 2015-09-08 | 2015-09-08 | Based on rough set and the method for diagnosing faults of the pump of BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510566043.0A CN106503798A (en) | 2015-09-08 | 2015-09-08 | Based on rough set and the method for diagnosing faults of the pump of BP neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106503798A true CN106503798A (en) | 2017-03-15 |
Family
ID=58287917
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510566043.0A Pending CN106503798A (en) | 2015-09-08 | 2015-09-08 | Based on rough set and the method for diagnosing faults of the pump of BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106503798A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107013449A (en) * | 2017-04-18 | 2017-08-04 | 山东万腾电子科技有限公司 | Voice signal based on deep learning recognizes the method and system of compressor fault |
| CN108763377A (en) * | 2018-05-18 | 2018-11-06 | 郑州轻工业学院 | Multi-source telemetering big data feature extraction preprocess method is diagnosed based on satellite failure |
| CN108921342A (en) * | 2018-06-26 | 2018-11-30 | 圆通速递有限公司 | A kind of logistics customer churn prediction method, medium and system |
| CN109034276A (en) * | 2018-09-04 | 2018-12-18 | 温州大学 | Valve body method for diagnosing faults based on BP neural network |
| CN109272123A (en) * | 2018-08-03 | 2019-01-25 | 常州大学 | It is a kind of based on convolution-Recognition with Recurrent Neural Network sucker rod pump operating condition method for early warning |
| CN109325470A (en) * | 2018-10-24 | 2019-02-12 | 山西潞安环保能源开发股份有限公司 | Working face in the pit homework type intelligent identification Method based on gas density parameter |
| CN110688809A (en) * | 2019-09-05 | 2020-01-14 | 西安理工大学 | Box transformer substation fault diagnosis method based on VPRS-RBF neural network |
| WO2020019681A1 (en) * | 2018-07-25 | 2020-01-30 | 北京国双科技有限公司 | Fault diagnosis method and apparatus for oil production equipment |
| CN117236173A (en) * | 2023-09-18 | 2023-12-15 | 中交四航局第一工程有限公司 | Subway station floor slab high formwork construction monitoring and safety early warning method |
| CN118690251A (en) * | 2024-08-22 | 2024-09-24 | 太原科技大学 | Plunger pump fault diagnosis method based on rough set and radial basis function neural network |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102254227A (en) * | 2011-07-06 | 2011-11-23 | 无锡泛太科技有限公司 | Rough set-based Back Propagation (BP) neural network fusion method of multiple sensors of Internet of things |
| CN104234695A (en) * | 2014-08-27 | 2014-12-24 | 陕西延长石油(集团)有限责任公司研究院 | Oil well fault diagnosis method based on neural network |
-
2015
- 2015-09-08 CN CN201510566043.0A patent/CN106503798A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102254227A (en) * | 2011-07-06 | 2011-11-23 | 无锡泛太科技有限公司 | Rough set-based Back Propagation (BP) neural network fusion method of multiple sensors of Internet of things |
| CN104234695A (en) * | 2014-08-27 | 2014-12-24 | 陕西延长石油(集团)有限责任公司研究院 | Oil well fault diagnosis method based on neural network |
Non-Patent Citations (2)
| Title |
|---|
| 刘卓 等: "基于BP神经网络和不变矩阵征特诊的泵功图诊断方法研究", 《制造业自动化》 * |
| 李坤杰: "粗糙集神经网络在注水泵机组故障诊断中的研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107013449A (en) * | 2017-04-18 | 2017-08-04 | 山东万腾电子科技有限公司 | Voice signal based on deep learning recognizes the method and system of compressor fault |
| CN108763377A (en) * | 2018-05-18 | 2018-11-06 | 郑州轻工业学院 | Multi-source telemetering big data feature extraction preprocess method is diagnosed based on satellite failure |
| CN108763377B (en) * | 2018-05-18 | 2021-08-13 | 郑州轻工业学院 | Multi-source telemetering big data feature extraction preprocessing method based on satellite fault diagnosis |
| CN108921342A (en) * | 2018-06-26 | 2018-11-30 | 圆通速递有限公司 | A kind of logistics customer churn prediction method, medium and system |
| CN108921342B (en) * | 2018-06-26 | 2022-07-12 | 圆通速递有限公司 | Logistics customer loss prediction method, medium and system |
| WO2020019681A1 (en) * | 2018-07-25 | 2020-01-30 | 北京国双科技有限公司 | Fault diagnosis method and apparatus for oil production equipment |
| CN109272123B (en) * | 2018-08-03 | 2021-06-22 | 常州大学 | Sucker-rod pump working condition early warning method based on convolution-circulation neural network |
| CN109272123A (en) * | 2018-08-03 | 2019-01-25 | 常州大学 | It is a kind of based on convolution-Recognition with Recurrent Neural Network sucker rod pump operating condition method for early warning |
| CN109034276A (en) * | 2018-09-04 | 2018-12-18 | 温州大学 | Valve body method for diagnosing faults based on BP neural network |
| CN109325470B (en) * | 2018-10-24 | 2021-09-03 | 山西潞安环保能源开发股份有限公司 | Intelligent underground working face operation type identification method based on gas concentration parameters |
| CN109325470A (en) * | 2018-10-24 | 2019-02-12 | 山西潞安环保能源开发股份有限公司 | Working face in the pit homework type intelligent identification Method based on gas density parameter |
| CN110688809A (en) * | 2019-09-05 | 2020-01-14 | 西安理工大学 | Box transformer substation fault diagnosis method based on VPRS-RBF neural network |
| CN117236173A (en) * | 2023-09-18 | 2023-12-15 | 中交四航局第一工程有限公司 | Subway station floor slab high formwork construction monitoring and safety early warning method |
| CN118690251A (en) * | 2024-08-22 | 2024-09-24 | 太原科技大学 | Plunger pump fault diagnosis method based on rough set and radial basis function neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106503798A (en) | Based on rough set and the method for diagnosing faults of the pump of BP neural network | |
| CN111738561B (en) | Regional water and soil loss dynamic monitoring method based on deep learning | |
| CN107133690B (en) | Optimal ordering method for river and lake water system communication engineering scheme | |
| KR102170765B1 (en) | Method for creating a shale gas production forecasting model using deep learning | |
| CN104572449A (en) | Automatic test method based on case library | |
| CN106257498A (en) | Zinc flotation work condition state division methods based on isomery textural characteristics | |
| CN101957889A (en) | Selective wear-based equipment optimal maintenance time prediction method | |
| CN106062311A (en) | Ranking drilling locations among shale plays | |
| CN112949196A (en) | Oil pumping well fault diagnosis method and system based on residual error neural network | |
| CN109763800A (en) | A kind of separated-zone water infection oil field amount prediction technique | |
| CN114357678B (en) | Novel optimization design method for regional underground water level monitoring network | |
| CN105447646A (en) | Health index assessment method for power distribution system | |
| CN104598766A (en) | Oil and gas reservoir comprehensive assessment oriented spatial case-based reasoning method | |
| CN104866940A (en) | Stochastic simulation method for geologic risk evaluation | |
| CN107330734A (en) | Business address system of selection based on Co location patterns and body | |
| Malo et al. | Eagle ford-introducing the big bad wolf | |
| RU2670801C9 (en) | System of integrated conceptual design of hydrocarbon fields | |
| Ariadji et al. | A novel tool for designing well placements by combination of modified genetic algorithm and artificial neural network | |
| CN115734274A (en) | Cellular network fault diagnosis method based on deep learning and knowledge graph | |
| Lu et al. | Identifying flow units by FA-assisted SSOM—An example from the Eocene basin-floor-fan turbidite reservoirs in the Daluhu Oilfield, Dongying Depression, Bohai Bay Basin, China | |
| CN106530109A (en) | Oilfield development appraisal well decision method based on information value | |
| Ariadji et al. | Optimization of vertical well placement for oil field development based on basic reservoir rock properties using a genetic algorithm | |
| CN111832951A (en) | Method and system for evaluating oil reservoir development value of small fault block ultra-low permeability reservoir | |
| CN117272841A (en) | Shale gas dessert prediction method based on hybrid neural network | |
| CN103049791A (en) | Training method of fuzzy self-organizing neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170315 |
|
| RJ01 | Rejection of invention patent application after publication |