CN109857979A - A kind of log approximating method based on wavelet analysis and BP neural network - Google Patents
A kind of log approximating method based on wavelet analysis and BP neural network Download PDFInfo
- Publication number
- CN109857979A CN109857979A CN201811641087.5A CN201811641087A CN109857979A CN 109857979 A CN109857979 A CN 109857979A CN 201811641087 A CN201811641087 A CN 201811641087A CN 109857979 A CN109857979 A CN 109857979A
- Authority
- CN
- China
- Prior art keywords
- neural network
- log
- layer
- wavelet analysis
- wavelet
- 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
Abstract
The present invention provides a kind of log approximating method based on wavelet analysis and BP neural network, include the following steps: that the log detected using wavelet analysis method to n probes independently carries out noise reduction process, isolates the useful signal of reflection formation characteristics;Establish BP neural network model, including input layer, hidden layer and output layer, choose the useful signal training BP neural network of n probe detection of several groups, the good BP neural network system of application training, it is that any time n is detected and through wavelet de-noising treated information data splitting as input vector, be input to the BP neural network of trained mistake, then obtain can correspondingly actual response stratum signature log curve.It is capable of the characteristic on actual response stratum using the obtained log of this method, and effect is good.
Description
Technical field
The invention belongs to technical field of radioactive measurement for wells, in particular to a kind of survey based on wavelet analysis and BP neural network
Well curve-fitting method.
Background technique
Log is the letter for being loaded with the various formation informations such as stratum micropore structure, fluid properties, lithology and lithofacies
Number.Four probe high-precision natural gamma logging devices use four probes independently, and can effectively obtain will be discharged in stratum
Gamma ray detection come out, to greatly improve the detection accuracy of natural gamma, the especially detection accuracy of thin and poor layer.
In four probe high-precision gamma ray logs, due to the random nature of nuclear decay, lead to each probe nature gal
Occur many statistic fluctuations unrelated with formation properties and burr interference on horse log, therefore to gamma ray log song
Before line analysis use, needs these statistic fluctuations and burr interference to be filtered, only retain reflection formation characteristics
Useful information.Due to the presence of statistic fluctuation error, prevent Natural Gamma-ray Logging Curves from completely, really reflect and detected
The real information arrived.Inhibit statistic fluctuation error, noise reduction process is done to log using wavelet de-noising, to a certain extent may be used
To eliminate influence of the statistic fluctuation error to curve, the real information of protection poor thin layer measurement.
Currently, the natural gamma information detected using four probes independently, forms a natural gamma
Log, it is main to be handled by digital filtering, the real information on a large amount of stratum will be lost in a sense.
The detection information for how making full use of four probes independently will reflect that the true useful signal of reservoir mentions
It takes out, and is integrated, ultimately forming one can be good at reflecting that reservoir especially thin and poor layer natural gamma rays are strong
The high-resolution Natural Gamma-ray Logging Curves of degree, are problems to be solved.
Summary of the invention
The present invention provides a kind of log approximating method based on wavelet analysis and BP neural network, utilizes this method
Obtained log is capable of the characteristic on actual response stratum, and effect is good.
To achieve the goals above, the invention is realized by the following technical scheme:
A kind of log approximating method based on wavelet analysis and BP neural network, includes the following steps:
Step 1, the log x that n probes independently are detected using wavelet analysis method1,...,xnInto
Row noise reduction process isolates the useful signal GR of reflection formation characteristics1,......,GRn;
Step 2 establishes BP neural network model, including input layer, hidden layer and output layer, and input layer has n node,
Hidden layer has m node, and output layer has 1 node;
Step 3, training BP neural network:
Choose several groupsAs input vector, input vector pair is obtained in code test well
The desired output vector GR answeredPhase, by GRiWith corresponding GRPhaseBP neural network system to be trained is inputted, with BP neural network
System output vector reality output GRIt is realWith desired output GRPhaseRoot-mean-square error as network performance function, BP in training process
Weight and threshold value in nerve network system are adjusted according to the network performance function of network, correct BP neural network system repeatedly
Unite obtained reality output GRIt is real, make it finally and GRPhaseRoot-mean-square error reach minimum;
Step 4 is being detected and handle through wavelet de-noising by any time n using BP neural network system
Information GR afterwards1,......,GRnData splitting is input to the BP neural network of trained mistake, then obtains phase as input vector
That answers is capable of the signature log curve on actual response stratum.
Further, the basic function of wavelet analysis described in step 1 of the present invention selects sym8 wavelet function, Decomposition order
It is 5 layers.
The utility model has the advantages that
BP neural network is applied to n probe high-precision Natural Gamma-ray Logging Curves fitting by the present invention, utilizes this method institute
The n probe Natural Gamma-ray Logging Curves of fitting and the curve of actual measurement are substantially overlapping, and the characteristic curve of fitting is smoother, fitting effect
Good, processing data precision is high.
The present invention uses wavelet analysis method before using neural metwork training, and the noise in sample has been carried out at noise reduction
Reason, wavelet analysis method have the good local character of time-domain and frequency domain, the noise reduction especially suitable for non-stationary signal
Useful signal in each probe can be decomposited and, so that the model of fit of BP neural network is more accurate by processing.
Detailed description of the invention
Fig. 1 is three layers of BP neural network structural model of the invention.
Specific embodiment
To make technical solution of the present invention and the relatively sharp protrusion of feature, further the present invention is carried out in conjunction with attached drawing detailed
Explanation.
Design philosophy of the invention: utilizing BP neural network, be fitted one it is stable, can really reflect reservoir rock characteristic
High-precision Natural Gamma-ray Logging Curves.Specifically: it is identical using 4 sizes, it the probe that is separated from each other while surveying
Gamma ray intensity is measured, 4 high-resolution Natural Gamma-ray Logging Curves are obtained;Due to being measured simultaneously there are statistic fluctuation error
Often there is larger difference in this obtained 4 GR logging curves, sufficiently reduce statistic fluctuation error using wavelet de-noising technology
Influence to measurement result, BP neural network are fitted 4 experiment curvs, obtain the relatively high natural gamma of resolution ratio
Log.
A kind of log approximating method based on wavelet analysis and BP neural network of the embodiment of the present invention, including walk as follows
It is rapid:
Step 1, using wavelet analysis method to four probe log x independently1,x2,x3,x4Carry out noise reduction
The useful signal GR of reflection formation characteristics is isolated in processing1,GR2,GR3,GR4。
Step 2 establishes BP neural network model, including input layer, hidden layer and output layer, and every layer includes several nodes,
It is not attached between every node layer.BP neural network structure is 4-9-1, i.e. input layer has 4 nodes, the input vector of input layer
GR1,GR2,GR3,GR4;Hidden layer is 1 hidden layer, has 9 nodes, output layer has 1 node, the output vector of output layer
GR。
Step 3, training BP neural network system, chooses several groupsAs input to
Amount, obtains the corresponding desired output vector GR of input vector in code test wellPhase, by GRiWith corresponding GRPhaseIt inputs wait instruct
Experienced BP neural network system, with BP neural network system output vector reality output GRIt is realWith desired output GRPhaseRoot mean square miss
Difference is used as network performance function, and the weight and threshold value in training process in BP neural network system are according to the error energy content of network
Number is adjusted, and corrects the reality output GR that BP neural network system obtains repeatedlyIt is real, make it finally and GRPhaseRoot-mean-square error
Reach minimum.
Step 4, using BP neural network system, by the GR of any time1,GR2,GR3,GR4Data splitting is as input
Vector is input to the BP neural network of trained mistake, is then exported accordingly.
The basic function of wavelet analysis described in step 1 selects sym8 wavelet function, and Decomposition order is 5 layers, to four difference
Independent probe log x1,x2,x3,x4Noise reduction process is carried out, it is many unrelated with formation properties occurring on log
Statistic fluctuation and burr interference are disposed, and the useful information GR of reflection formation characteristics is only retained1,GR2,GR3,GR4, convenient for BP mind
Training through network.
It, only need to be according to system since BP neural network matched curve does not need that the equation form to matched curve is known in advance
Input value and its corresponding output valve can be fitted, BP neural network model have is forced compared with strong nonlinearity processing capacity and function
Nearly ability, network operations speed is fast, and performance is stablized.
In conclusion the above is merely preferred embodiments of the present invention, being not intended to limit the scope of the present invention.
All within the spirits and principles of the present invention, any modification, equivalent replacement, improvement and so on should be included in of the invention
Within protection scope.
Claims (2)
1. a kind of log approximating method based on wavelet analysis and BP neural network, which comprises the steps of:
Step 1, the log x that n probes independently are detected using wavelet analysis method1,...,xnIt is dropped
It makes an uproar processing, isolates the useful signal GR of reflection formation characteristics1,......,GRn;
Step 2 establishes BP neural network model, including input layer, hidden layer and output layer, and input layer has n node, implies
Layer has m node, and output layer has 1 node;
Step 3, training BP neural network:
Choose several groupsAs input vector, it is corresponding in code test well to obtain input vector
Desired output vector GRPhase, by GRiWith corresponding GRPhaseBP neural network system to be trained is inputted, with BP neural network system
Output vector reality output GRIt is realWith desired output GRPhaseRoot-mean-square error as network performance function, BP nerve in training process
Weight and threshold value in network system are adjusted according to the network performance function of network, are corrected BP neural network system repeatedly and are obtained
The reality output GR arrivedIt is real, make it finally and GRPhaseRoot-mean-square error reach minimum;
Step 4 is being detected and treated through wavelet de-noising by any time n using BP neural network system
Information GR1,......,GRnData splitting is input to the BP neural network of trained mistake as input vector, then obtains corresponding
It is capable of the signature log curve on actual response stratum.
2. the log approximating method based on wavelet analysis and BP neural network according to claim 1, which is characterized in that
The basic function of wavelet analysis described in step 1 selects sym8 wavelet function, and Decomposition order is 5 layers.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811641087.5A CN109857979A (en) | 2018-12-29 | 2018-12-29 | A kind of log approximating method based on wavelet analysis and BP neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811641087.5A CN109857979A (en) | 2018-12-29 | 2018-12-29 | A kind of log approximating method based on wavelet analysis and BP neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109857979A true CN109857979A (en) | 2019-06-07 |
Family
ID=66893394
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811641087.5A Pending CN109857979A (en) | 2018-12-29 | 2018-12-29 | A kind of log approximating method based on wavelet analysis and BP neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109857979A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110552693A (en) * | 2019-09-19 | 2019-12-10 | 中国科学院声学研究所 | layer interface identification method of induction logging curve based on deep neural network |
CN111753958A (en) * | 2020-06-22 | 2020-10-09 | 成都理工大学 | Lamp shade group microorganism rock microphase identification method based on logging data deep learning |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1799013A (en) * | 2003-06-05 | 2006-07-05 | 罗斯蒙德公司 | Process device diagnostics using process variable sensor signal |
CN103499345A (en) * | 2013-10-15 | 2014-01-08 | 北京航空航天大学 | Fiber-optic gyro temperature drift compensating method based on wavelet analysis and BP (back propagation) neutral network |
CN104146693A (en) * | 2014-07-30 | 2014-11-19 | 天津大学 | Wavelet transform and curve fitting based pulse wave preprocessing method |
CN108171159A (en) * | 2017-12-27 | 2018-06-15 | 山东大学 | A kind of measurement while drilling signal processing method based on Wavelet Denoising Method and neural network recognization |
-
2018
- 2018-12-29 CN CN201811641087.5A patent/CN109857979A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1799013A (en) * | 2003-06-05 | 2006-07-05 | 罗斯蒙德公司 | Process device diagnostics using process variable sensor signal |
CN103499345A (en) * | 2013-10-15 | 2014-01-08 | 北京航空航天大学 | Fiber-optic gyro temperature drift compensating method based on wavelet analysis and BP (back propagation) neutral network |
CN104146693A (en) * | 2014-07-30 | 2014-11-19 | 天津大学 | Wavelet transform and curve fitting based pulse wave preprocessing method |
CN108171159A (en) * | 2017-12-27 | 2018-06-15 | 山东大学 | A kind of measurement while drilling signal processing method based on Wavelet Denoising Method and neural network recognization |
Non-Patent Citations (2)
Title |
---|
李瑜等: "神经网络方法拟合核磁共振测井曲线", 《微电子学与计算机》 * |
郭锐: "测井数据的小波分析方法", 《中国优秀博硕士学位论文全文数据库(硕士) 基础科学辑》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110552693A (en) * | 2019-09-19 | 2019-12-10 | 中国科学院声学研究所 | layer interface identification method of induction logging curve based on deep neural network |
CN111753958A (en) * | 2020-06-22 | 2020-10-09 | 成都理工大学 | Lamp shade group microorganism rock microphase identification method based on logging data deep learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113759425B (en) | Method and system for evaluating filling characteristics of deep paleo-karst reservoir stratum by well-seismic combination | |
CN109425896B (en) | Dolomite oil and gas reservoir distribution prediction method and device | |
CN111239802B (en) | Deep learning speed modeling method based on seismic reflection waveform and velocity spectrum | |
CN113759424B (en) | Karst reservoir filling analysis method and system based on spectral decomposition and machine learning | |
WO2017084454A1 (en) | Stratum component optimization determination method and device | |
CN106405651B (en) | Full waveform inversion initial velocity model construction method based on logging matching | |
Yang et al. | Simultaneous earthquake detection on multiple stations via a convolutional neural network | |
CN104280777B (en) | Method for suppressing interference of seismic data multiples on land | |
CN102353985A (en) | Pseudo-acoustic curve construction method based on nonsubsampled Contourlet transformation | |
CN107942404B (en) | A kind of method and device in determining crack | |
CN108693555A (en) | Intelligent time-varying blind deconvolution wideband processing method and processing device | |
CN105607122B (en) | A kind of earthquake texture blending and Enhancement Method based on full variation geological data decomposition model | |
CN109799530A (en) | Rayleigh waves dispersion curve inversion method for seismic surface wave exploration | |
CN104316966B (en) | A kind of Fluid Identification Method and system | |
Si et al. | Attenuation of random noise using denoising convolutional neural networks | |
CN104849763A (en) | Method for standardizing multi-well logging curves with low-frequency trends | |
CN109857979A (en) | A kind of log approximating method based on wavelet analysis and BP neural network | |
CN110454153A (en) | A kind of nuclear magnetic resonance log relaxation inversion method | |
CN104730572B (en) | A kind of diffracted wave formation method based on L0 semi-norm and device | |
RU2490677C2 (en) | Method for complex processing of geophysical data "litoscan" system for realising said method | |
CN110118994B (en) | Continental facies hydrocarbon source rock quantitative prediction method based on seismic inversion and machine learning | |
CN107526117A (en) | SVEL Forecasting Methodology based on autocoding and the study joint network that transfinites | |
CN105445795B (en) | The velocity analysis method and its system of tilted interface | |
CN110658558A (en) | Front-of-stack depth reverse time migration imaging method and system for absorption attenuation medium | |
Biryol et al. | Imaging the shallow crust with local and regional earthquake tomography |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190607 |