CN103049791A - Training method of fuzzy self-organizing neural network - Google Patents
Training method of fuzzy self-organizing neural network Download PDFInfo
- Publication number
- CN103049791A CN103049791A CN 201110326271 CN201110326271A CN103049791A CN 103049791 A CN103049791 A CN 103049791A CN 201110326271 CN201110326271 CN 201110326271 CN 201110326271 A CN201110326271 A CN 201110326271A CN 103049791 A CN103049791 A CN 103049791A
- Authority
- CN
- China
- Prior art keywords
- training
- neural network
- seismic
- organizing neural
- network
- 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
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a training method of a fuzzy self-organizing neural network, which comprises the steps of: a, determining a training sample x; b, randomly initializing a weight value wij, wherein the wij is more than 0 and less than 1, i is equal to 0, 1, ellipsis, N-1, and j is equal to 0, 1, ellipsis, K-1; c, inputting all sample points, calculating a membership degree of each sample to all subsets; d, regulating a network weight; and e, according to the judging condition of network stabilization, if the judging condition is met, ending the learning, and if the judging condition is not met, transferring into the step c for continuously learning. The training method is capable of completing the training of the self-organizing neutral network, and is good in training effect and low in training cost.
Description
Technical field
The present invention relates to a kind of training method of Fuzzy Self-organizing Neural Network.
Background technology
Along with the development of Discussion of Earthquake Attribute Technology, Seismic Reservoir Prediction has become the effective means that instructs oil-gas exploration and development.Yet because seismic properties is of a great variety, and the relation between the forecasting object is complicated, and different work areas and different reservoir are incomplete same to (the most effective, the most representative) seismic properties of institute's target of prediction sensitivity.Even same work area, same reservoir, forecasting object is different, and corresponding Sensitive Attributes also there are differences.
Because this multi-solution of seismic properties, so that some attribute can have a strong impact on the precision of reservoir prediction, therefore seismic properties is in optimized selection and just seems very necessary.Optimum methods of seismic attributes can significantly improve the precision of Seismic Reservoir Prediction, more effectively carries out reservoir and describes, and further improves the drilling well success ratio, has obvious economic benefit and social benefit.
Because seismic properties refers to that by prestack or post-stack seismic data some that derive through mathematic(al) manipulation comprise that outside geometric shape, internal reflection structure, continuity, amplitude, frequency and speed etc. represent the parameter of characteristics of seismic.And the seismic reflection unit in the three dimensions that seismic facies is specific seismic reflection parameter to be limited, it is the seismic response of particular deposition phase or geologic body.Therefore, it is very significant using seismic properties division seismic facies type.At last, explain sedimentary facies and the sedimentary environment of these seismic facies representatives by seismic facies analysis, be converted to the purpose of sedimentary facies to reach seismic facies.
At present, the normal self organizing neural network that adopts carries out cluster analysis to reach the purpose of dividing the seismic facies type to seismic properties, but traditional self organizing neural network just is grouped in the nearest class subset owing to training sample of every input, this training patterns may be too hurry, affect network to the grasp of all training sample features, and then the correctness of impact classification.Also very easily cause simultaneously the vibration of network weight, so that learning time is longer.In addition, the choosing of network parameter such as the gain function of self organizing neural network, bounding function, neighborhood are very stubborn problems, and they change along with the difference of drawing number of categories.In view of these problems of self organizing neural network, the mode that this paper adopts a kind of Fuzzy Self-organizing Neural Network to be combined with seismic properties is carried out SEISMIC PHASE BY PATTERN RECOGNITION.Fuzzy Self-organizing Neural Network is different from traditional self organizing neural network, and it is once to input all training sample points, determines that each sample point is to the subjection degree of every class subset.The adjustment of network weight has considered the characteristic information of all samples, and one takes turns study only adjusts once, has greatly saved learning time.And this method carries out seismic facies analysis, can effectively identify Sedimentary facies and the geological phenomenons such as river course, delta, outfall fan, tomography, lithologic anomalous body, forms a kind of practical, reservoir prediction technique that precision is high.
Seismic properties optimization be exactly optimize the most responsive to Solve problems, the most effectively or the attribute of representative is arranged most, in order to improve the precision of reservoir prediction.Before carrying out the seismic properties optimization process, usually to carry out to all properties that extracts standardization (such as normalization etc.).The optimization of seismic properties starts from " bright spot " technology that 20 century 70s occur, and in this technology, selects reflection wave amplitude and polarity etc., i.e. early stage " expert's optimization ".Along with the development of artificial intelligence technology, most methods are introduced in the optimization method of seismic properties.
At present the optimization method of attribute is more, but it can be divided into two large classes: utilize expertise to be optimized and utilize mathematical method to carry out Automatic Optimal.Expert method can not satisfy the requirement of present reservoir prediction, can only be as a kind of auxiliary means.The outer seismic properties method for optimizing of Present Domestic mainly is mathematical method, mainly contains Karhunen-Loeve transformation, local linear embedding algorithm (LLE), Isometric Maps (ISOMAP), multiple discriminant analysis method (MDA), attribute contribution amount method, searching algorithm, genetic algorithm, Rough Set (RS) etc.
Along with the development of Discussion of Earthquake Attribute Technology, Earthquake Reservoir has also obtained faster development as a branch.Utilize finally multiple seismic properties to carry out the technology of reservoir prediction from early stage single attribute forecast; From early stage expert method artificial intelligence approach finally.From the eighties, " pattern-recognition " is subject to special attention, successively worked out " Fuzzy Pattern Recognition ", " statistical model identification ", " network mode identification " and methods such as " approximation of function ", reservoir prediction technique has obtained fast development after this.Forecasting object develops into predicting reservoir parameter and formation lithology etc. from predicting oil/gas.At present, can be divided into according to Forecasting Methodology: the prediction of approximation of function class and the prediction of pattern-recognition class.The approximation of function class methods mainly are that reservoir parameter etc. is predicted that major parameter comprises Sandstone Percentage, factor of porosity, oil saturation, reservoir thickness, reservoir pressure etc., often adopt BP neural network, radial base neural net, CUSI network etc.The pattern-recognition class methods are mainly used in oil and gas prediction, SEISMIC PHASE BY PATTERN RECOGNITION, and the method for employing is transitioned into self organizing neural network, BP neural network, fractal theory, gray theory etc. from statistical model identification, Fuzzy Pattern Recognition.
The seismic facies analysis technology has also obtained very fast development naturally as the part of reservoir prediction.It is a kind of geological method that utilizes seismic data to carry out geologic interpretation that grows up late 1970s.Development so far, seismic facies analysis by naked eyes judge seismic facies unit various parameters, make the seismic facies map by hand to the seismic facies parameter of self organizing neural network judgement different units, and directly the seismic facies parameter is classified.Initial manual operations, time-consuming taking a lot of work, particularly when reflectance anomaly on the earthquake section was not outstanding, this work was more difficult, had developed into afterwards with statistical model identification and fuzzy clustering and had automatically divided seismic facies.But statistical model identification is high to the requirement of attributes extraction and selection, can only be applicable to several simple forms, fuzzy clustering method is in that to set up accurately reasonably aspect the membership function difficulty larger, and when data volume is large operation time long, sometimes can realize hardly.Use afterwards nerual network technique to carry out pattern-recognition and obtained good effect.Because artificial neural network can be processed some circumstance complications, unclear, the indefinite problem of inference rule of background knowledge, and allow sample that larger damaged and distortion is arranged.Just at present about dividing the article of seismic facies, what multiselect was got is the Kohonene self organizing neural network.
One of key of pattern-recognition is not only in seismic properties optimization, and is also significant to improving approximation of function method Seismic Reservoir Prediction precision.In Seismic Reservoir Prediction, usually extract a plurality of attributes, adopt pattern-recognition or approximation of function method to carry out reservoir prediction.But in different regions, the different layers position, be incomplete same to (or effective, most representative) seismic properties of institute's forecasting object sensitivity; Even in areal, same layer position, also be discrepant to the seismic properties of the object sensitivity predicted.Therefore be necessary the optimum methods of seismic attributes in the Study In Reservoir prediction.
At present the optimization method of seismic properties is more, but it can be divided into two large classes: utilize expertise to be optimized and utilize mathematical method to carry out Automatic Optimal.The expert optimizes, and in general the oil field expert knows quite well with the seismic properties of maximum reservoir information certain area, can carry out by rule of thumb seismic properties and select.Sometimes the expert can propose several groups of more excellent attributes or combinations of attributes, but which is organized optimum difficulty and draws a conclusion.This can by calculate misclassification rate (pattern-recongnition method) or predicated error (approximation of function method) and comparing, choose the little person of misclassification rate or predicated error and be optimum seismic properties or seismic properties combination.Compare with expert's optimization method, Mathematics Optimization Method is more complex, and has widely applicability.
Rough set (rough set is called for short RS or rough set) theory is nineteen eighty-two, and the Z.Pawlak professor of Warsaw, POL Polytechnics is take relational theory as basic tool, promotes that traditional sets theory puts forward, and has published the rough set theory monograph in 1991.For the research deficiency of data is analyzed, reasoning, find the relation between data, extract useful attribute, simplify information processing, research out of true, the expression of uncertain knowledge, study, inductive method provide a strong instrument.Simultaneously, the RS theory also provides new logic of science and research method for information science and cognitive science, and provides effective treatment technology for Intelligent Information Processing.The RS theory need not to provide the outer any prior imformation of data acquisition of the required processing of problem, and only data are deleted redundant information according to the observation, more incomplete knowledge level---roughness, the dependence between attribute and importance, the ability of extraction classifying rules etc.Thus, rough set is knowledge discovery in database, expert system, decision support system (DSS), pattern-recognition, fuzzy control etc., and a kind of new mathematical method is provided.
Summary of the invention
Purpose of the present invention is in order to overcome the deficiencies in the prior art and defective, a kind of training method of Fuzzy Self-organizing Neural Network is provided, the training method of this Fuzzy Self-organizing Neural Network can be finished the training to self organizing neural network, and training effect is good, and the training cost is low.
Purpose of the present invention is achieved through the following technical solutions: the training method of Fuzzy Self-organizing Neural Network may further comprise the steps:
(a) determine training sample x;
(b) random initializtion weight w
Ij, 0＜w
Ij＜1, i=0,1 ..., N-1; J=0,1 ..., K-1;
(c) input all sample points, calculate each sample to the degree of membership of all subsets;
(d) adjust network weight;
(e) according to the decision condition of network stabilization, as satisfying, then study finishes, and as not satisfying, then changes step (c) continue studying over to.
Described study factor index a is 1 or 2.
Described training sample number is M, and characteristic number is N, and number of categories is K.
In sum, the invention has the beneficial effects as follows: can finish the training to self organizing neural network, and training effect is good, the training cost is low.
Embodiment
Below in conjunction with embodiment, to the detailed description further of the present invention's do, but embodiments of the present invention are not limited to this.
Embodiment:
Present embodiment relates to the training method of Fuzzy Self-organizing Neural Network, may further comprise the steps:
(a) determine training sample x;
(b) random initializtion weight w
Ij, 0＜w
Ij＜1, i=0,1 ..., N-1; J=0,1 ..., K-1;
(c) input all sample points, calculate each sample to the degree of membership of all subsets;
(d) adjust network weight;
(e) according to the decision condition of network stabilization, as satisfying, then study finishes, and as not satisfying, then changes step (c) continue studying over to.
Described study factor index a is 1 or 2.
Described training sample number is M, and characteristic number is N, and number of categories is K.
The above only is preferred embodiment of the present invention, is not the present invention is done any pro forma restriction, and any simple modification, the equivalent variations on every foundation technical spirit of the present invention above embodiment done all fall within protection scope of the present invention.
Claims (3)
1. the training method of Fuzzy Self-organizing Neural Network is characterized in that, may further comprise the steps:
(a) determine training sample x;
(b) random initializtion weight w
Ij, 0＜w
Ij＜1, i=0,1 ..., N-1; J=0,1 ..., K-1;
(c) input all sample points, calculate each sample to the degree of membership of all subsets;
(d) adjust network weight;
(e) according to the decision condition of network stabilization, as satisfying, then study finishes, and as not satisfying, then changes step (c) continue studying over to.
2. the training method of Fuzzy Self-organizing Neural Network according to claim 1 is characterized in that, described study factor index a is 1 or 2.
3. the training method of Fuzzy Self-organizing Neural Network according to claim 1 is characterized in that, described training sample number is M, and characteristic number is N, and number of categories is K.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110326271 CN103049791A (en) | 2011-10-13 | 2011-10-13 | Training method of fuzzy self-organizing neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110326271 CN103049791A (en) | 2011-10-13 | 2011-10-13 | Training method of fuzzy self-organizing neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103049791A true CN103049791A (en) | 2013-04-17 |
Family
ID=48062423
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 201110326271 Pending CN103049791A (en) | 2011-10-13 | 2011-10-13 | Training method of fuzzy self-organizing neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103049791A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104453875A (en) * | 2014-10-29 | 2015-03-25 | 中国石油集团川庆钻探工程有限公司 | Shale gas reservoir identification method based on self-organizing competitive neural network |
| CN105549384A (en) * | 2015-09-01 | 2016-05-04 | 中国矿业大学 | Inverted pendulum control method based on neural network and reinforced learning |
| CN106920007A (en) * | 2017-02-27 | 2017-07-04 | 北京工业大学 | PM based on second order Self-organized Fuzzy Neural Network2.5Intelligent Forecasting |
| CN109613825A (en) * | 2018-12-13 | 2019-04-12 | 北京北排科技有限公司 | Sewage treatment plant's intelligent patrol detection track antidote based on Self-organized Fuzzy Neural Network |
-
2011
- 2011-10-13 CN CN 201110326271 patent/CN103049791A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104453875A (en) * | 2014-10-29 | 2015-03-25 | 中国石油集团川庆钻探工程有限公司 | Shale gas reservoir identification method based on self-organizing competitive neural network |
| CN104453875B (en) * | 2014-10-29 | 2018-05-01 | 中国石油集团川庆钻探工程有限公司 | Shale gas reservoir identification method based on self-organizing competitive neural network |
| CN105549384A (en) * | 2015-09-01 | 2016-05-04 | 中国矿业大学 | Inverted pendulum control method based on neural network and reinforced learning |
| CN105549384B (en) * | 2015-09-01 | 2018-11-06 | 中国矿业大学 | A kind of inverted pendulum control method based on neural network and intensified learning |
| CN106920007A (en) * | 2017-02-27 | 2017-07-04 | 北京工业大学 | PM based on second order Self-organized Fuzzy Neural Network2.5Intelligent Forecasting |
| CN106920007B (en) * | 2017-02-27 | 2020-07-17 | 北京工业大学 | PM based on second-order self-organizing fuzzy neural network2.5Intelligent prediction method |
| CN109613825A (en) * | 2018-12-13 | 2019-04-12 | 北京北排科技有限公司 | Sewage treatment plant's intelligent patrol detection track antidote based on Self-organized Fuzzy Neural Network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109800863B (en) | Logging phase identification method based on fuzzy theory and neural network | |
| CN112083498B (en) | Multi-wave earthquake oil and gas reservoir prediction method based on deep neural network | |
| CN107545577B (en) | Sedimentary facies image segmentation method based on neural network | |
| CN104570083B (en) | Geologic body automatic identifying method based on multi-dimensional earthquake attribute | |
| CN112613536B (en) | Near infrared spectrum diesel fuel brand recognition method based on SMOTE and deep learning | |
| CN110346831B (en) | Intelligent seismic fluid identification method based on random forest algorithm | |
| CN102622515B (en) | A kind of weather prediction method | |
| CN109184677A (en) | Reservoir evaluation methods for heterogeneous alternating layers sand body | |
| Zhu et al. | Rapid identification of high-quality marine shale gas reservoirs based on the oversampling method and random forest algorithm | |
| CN105760673A (en) | Fluvial facies reservoir earthquake sensitive parameter template analysis method | |
| CN112614021B (en) | Tunnel surrounding rock geological information prediction method based on built tunnel information intelligent identification | |
| CN106503798A (en) | Based on rough set and the method for diagnosing faults of the pump of BP neural network | |
| CN103049791A (en) | Training method of fuzzy self-organizing neural network | |
| CN112576238B (en) | System, method and application for determining position and content of residual oil in low-permeability reservoir | |
| CN103376468A (en) | Reservoir parameter quantitative characterization method based on neural network function approximation algorithm | |
| CN106446514A (en) | Fuzzy theory and neural network-based well-log facies recognition method | |
| CN106569272A (en) | Earthquake attribute fusion method based on data property space ascending dimension | |
| CN103048688A (en) | Seismic attribute optimal selection method based on three-step method | |
| CN117272841A (en) | Shale gas dessert prediction method based on hybrid neural network | |
| CN103049463A (en) | Seismic attribute optimal selection method based on rough set theory | |
| CN115660221B (en) | Oil and gas reservoir economic recoverable reserve assessment method and system based on hybrid neural network | |
| CN117093922A (en) | Improved SVM-based complex fluid identification method for unbalanced sample oil reservoir | |
| CN103048687A (en) | Novel seismic attribute selecting method | |
| CN103049632A (en) | Non-supervision dimension reducing method aiming at nonlinear data | |
| CN103048686A (en) | Method for quickly solving seismic attribute reduction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20130417 |