CN109670533A - A kind of multiple dimensioned well-log facies recognition method based on convolutional neural networks - Google Patents

A kind of multiple dimensioned well-log facies recognition method based on convolutional neural networks Download PDF

Info

Publication number
CN109670533A
CN109670533A CN201811412931.7A CN201811412931A CN109670533A CN 109670533 A CN109670533 A CN 109670533A CN 201811412931 A CN201811412931 A CN 201811412931A CN 109670533 A CN109670533 A CN 109670533A
Authority
CN
China
Prior art keywords
layer
data
convolutional neural
neural networks
model
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
Application number
CN201811412931.7A
Other languages
Chinese (zh)
Inventor
何旭
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Individual
Original Assignee
Individual
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Individual filed Critical Individual
Priority to CN201811412931.7A priority Critical patent/CN109670533A/en
Publication of CN109670533A publication Critical patent/CN109670533A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Abstract

The multiple dimensioned well-log facies recognition method based on convolutional neural networks that the invention proposes a kind of, comprising the following steps: step (a) acquires initial data；Step (b) establishes training dataset；Step (c) constructs the well logging phase disaggregated model based on convolutional neural networks, identifies to data to be identified, and the model includes 8 layer networks, wherein the 1st layer, the 3rd layer and the 5th layer is convolutional layer, for extracting feature；2nd layer and the 4th layer is pond layer, for reducing model complexity；6th layer and the 7th layer is full articulamentum, for two dimensional character is changed into one-dimensional characteristic vector；8th layer is output layer, for classifying.The present invention effectively limits the complexity of model using L2 norm, dropout and LRN, improve model generalization ability, the category of model accuracy rate compared to the 97.87% of other algorithms is obtained, so as to more accurately automatic identification form of logs.

Description

A kind of multiple dimensioned well-log facies recognition method based on convolutional neural networks
Technical field
The present invention relates to well-log facies recognition field, in particular to a kind of multiple dimensioned well logging acquaintance based on convolutional neural networks Other method.
Background technique
Well logging is to be connected the medium scale information of rock core and earthquake, and recognize the main approach of well point geological information, Since the concept for proposing well logging phase from Schlumberger the 1950s well logging expert, how well logging phase has that accurately and quickly been identified An important process as reservoir geology research.Early stage well logging is mutually studied mainly by combining form of logs and well logging to provide Comparative analysis is expected, so that it is determined that sedimentary facies or stratigraphic sequence, are a kind of time-consuming and laborious full manual analysis modes.With computer The development of technology, especially artificial intelligence, make it possible log well phase it is accurate, fast and automatically identify, techniques of discriminant analysis, number The methods of statistic law, clustering methodology, fuzzy diagnosis, curve matching, support vector machines and neural network is managed gradually to be applied to survey Well is mutually identified and is analyzed.
But it is above-mentioned using machine learning algorithm carry out the model of well logging phase quantitative judge analysis there is also it is many not Foot: firstly, feature extraction is cumbersome, the operation such as need for data to be normalized, different method for normalizing is to model training knot Fruit also has an impact, and the data selected when being identified also need to carry out dimension adjustment；Secondly, generalization ability is insufficient, this The model that kind quantitative analysis method training obtains is only applicable to other areas of this area's either same deposition environment.
Most begin to early in the eighties in last century it has been proposed that convolutional neural networks (Convolutional Neural Network, CNN) and achieve in Handwritten Digit Recognition excellent achievement.The structure of CNN is mainly by input layer, convolution Layer, pond layer (also referred to as down-sampling layer), full articulamentum and output layer are constituted, convolutional layer building adjacent with pond layer, according to reality It needs can be set multiple.
What convolutional layer can be considered as being made of multiple Feature Mappings (Feature Map), multiple neuron composition characteristics reflect It penetrates, each neuron is connected by convolution kernel (Kernel) with a part of neuron of upper layer Feature Mapping.Wherein convolution kernel is The weight matrix of one random initializtion, effect are to carry out following convolution operation with the Feature Mapping of this layer
Wherein, n × n is convolution kernel size, xiIt is the input value of ith pixel point in region, wiIt is corresponding weight matrix Value, b are the biasings being added after calculating, and the value that last convolution operation obtains this region is y.
Convolution operation operates entire Feature Mapping using convolution kernel, mainly enables to weight shared, subtracts significantly Few network parameter, this is also a big advantage of CNN.Convolution terminates using excitation function by y normalization, as next layer of input
X=f (y)
F () is excitation function, and there are many forms, according to circumstances can voluntarily be selected in actual experiment, and x is excitation letter The input value of several output and next layer network.
Pond layer is also to be made of multiple Feature Mappings, and Feature Mapping is one-to-one with upper one layer].With maximum pond Turn to example
yi=max { x1, x2..., xn×n}
Wherein, x is the value of each point in pond region, yiFor the maximum value chosen from n × n point, that is, pond The output of operation.
Pondization, which operates, reduces the resolution ratio of Feature Mapping, so that the feature with translation, invariable rotary shape is obtained, and Reduce neuron number, to reduce the complexity of model, improves generalization ability.
Therefore, how convolutional neural networks are based on, a kind of method for capableing of effective Classification and Identification well logging phase is provided, is current Urgent problem to be solved.
Summary of the invention
The present invention proposes a kind of multiple dimensioned well-log facies recognition method based on convolutional neural networks, being capable of effective Classification and Identification Well logging phase, is better than other algorithms in nicety of grading and model generalization ability.
The technical scheme of the present invention is realized as follows:
A kind of multiple dimensioned well-log facies recognition method based on convolutional neural networks, comprising the following steps:
Step (a) acquires initial data.
Step (b) establishes training dataset.
Step (c) constructs the well logging phase disaggregated model based on convolutional neural networks, identifies to data to be identified, institute Stating model includes 8 layer networks, wherein the 1st layer, the 3rd layer and the 5th layer is convolutional layer, for extracting feature；2nd layer and the 4th layer is Pond layer, for reducing model complexity；6th layer and the 7th layer is full articulamentum, for two dimensional character is changed into one-dimensional characteristic Vector；8th layer is output layer, for classifying.
Optionally, convolution operation is carried out to input picture using convolution kernel at the 1st layer, then uses ReLU activation primitive pair Each click-through line activating processing, ReLU function formula are as follows:
F (x)=max (0, x)
Wherein, x is the point value, and f (x) is the new numerical value obtained by activation primitive.
Optionally, the method also includes: joined in the model and to lose data, local acknowledgement's normalization and L2 just It is then tactful.
Optionally, the method also includes: select Natural Gamma-ray Logging Curves form as feature, by the original number of acquisition According to being changed into image format.
Optionally, the method also includes: be first threshold value with 120, find out GR numerical value be higher than 120 extreme point, will count It is primary according to segmentation；Then it is threshold value with 100, finds out the extreme value between previous segmentation section greater than 100, secondary point is carried out to data It cuts；It is again threshold value with 85, finds out the extreme value between previous segmentation section greater than 85, data are divided three times；Finally by these Data segment is converted into 2-D gray image form.
Optionally, the method also includes: select Daubechies wavelet basis to log data carry out different scale it is small Wave conversion.
The beneficial effects of the present invention are:
(1) complexity that model is effectively limited using L2 norm, dropout and LRN, improves model generalization energy Power obtains the category of model accuracy rate compared to the 97.87% of other algorithms, so as to more accurately automatic identification Form of logs.
It (2) due to method selection of the invention is that the well logging of two-dimensional image data is mutually used as mode to classify, it is unrelated Specific value, that is, combine to be analyzed area depositional environment background method proposed by the invention can be used to log well phase divide Analysis, so having generalization ability more preferably than classical way.
(3) it is divided using the well logging phase element that Daubechies small echo carries out different scale, becomes well-log facies recognition work Obtain more convenient and efficient.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with It obtains other drawings based on these drawings.
Fig. 1 is a kind of flow chart of the multiple dimensioned well-log facies recognition method based on convolutional neural networks of the present invention；
Fig. 2 is the schematic diagram of well logging phase disaggregated model of the invention；
Fig. 3 is Sigmoid and Tanh activation primitive waveform diagram；
Fig. 4 is ReLU activation primitive waveform diagram；
Fig. 5 is the morphological feature curve synoptic diagram in log data；
Fig. 6 is the comparison schematic diagram of raw-data map with the datagram after wavelet transform process.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other Embodiment shall fall within the protection scope of the present invention.
As shown in Figure 1, the invention proposes a kind of multiple dimensioned well-log facies recognition method based on convolutional neural networks, including Following steps:
Step (a) acquires initial data.
Step (b) establishes training dataset.
Step (c) constructs the well logging phase disaggregated model based on convolutional neural networks, identifies to data to be identified, institute Stating model includes 8 layer networks, as shown in Figure 2, wherein the 1st layer, the 3rd layer and the 5th layer is convolutional layer, for extracting feature；2nd Layer is pond layer with the 4th layer, for reducing model complexity；6th layer and the 7th layer is full articulamentum, is used to turn two dimensional character Become one-dimensional characteristic vector；8th layer is output layer, for classifying.
Optionally, convolution operation is carried out to input picture using convolution kernel at the 1st layer, then uses ReLU activation primitive pair Each click-through line activating processing, ReLU function formula are as follows:
F (x)=max (0, x)
Wherein, x is the point value, and f (x) is the new numerical value obtained by activation primitive.
For example, input picture size is 227 × 227 pixels, in the 1st layer of convolution kernel using 64 11 × 11 sizes to defeated Enter image and carries out convolution operation.
There are many activation primitives, generally uses saturation nonlinearity function, such as Sigmoid activation primitive or Tanh activation Function etc., and ReLU is unsaturated nonlinear function, as shown in Figure 3 and Figure 4, abscissa indicates the input value of activation primitive, is Real number (no unit), ordinate indicate the output valve obtained by activation primitive operation, are compared to Sigmoid and Tanh activation Function, ReLU activation primitive can be easier to acquire its first derivative when carrying out gradient decline, model can be made to be less prone to Gradient disappears or explosion, moreover it is possible to accelerate its convergence rate, so this model selection ReLU is as activation primitive；
Optionally, the input in the Feature Mapping obtained after one layer of convolution as pond layer, the 2nd layer uses pond It checks it and carries out pondization operation.Common pond method has maximum pond and average pond, that is, takes the maximum in the part Chi Huahe The average value of value and all values.For the characteristic of established data set, i.e., entire picture is largely blank parts, is really had The characteristic area accounting very little of value, model selection maximum pondization operation take full advantage of that it is suitable for separating sparse spy The characteristics of sign.
For example, its progress pondization of the 2nd layer of pondization verification using 192 5 × 5 sizes operates.
Optionally, the 3rd layer and the 5th layer is both convolutional layer, has used convolution kernel to be handled respectively, has then reused ReLU Activation primitive processing.
Optionally, pondization verification Feature Mapping has been used to carry out maximum pondization processing for the 4th layer.
Optionally, the Feature Mapping that front convolution is completed is transformed into feature vector the 6th layer.
Optionally, primary full attended operation is carried out again to the 6th layer of result the 7th layer.
For example, input picture size is 227 × 227 pixels, firstly, in the 1st layer of convolution using 64 11 × 11 sizes It checks input picture and carries out convolution operation, then each click-through line activating is handled using ReLU activation primitive, ReLU function is public Formula is as follows:
F (x)=max (0, x)
Wherein, x is the point value, and f (x) is the new numerical value obtained by activation primitive.
Then, the input in the Feature Mapping obtained after one layer of convolution as pond layer, the 2nd layer uses 192 5 Its progress pondization operation of the pondization verification of × 5 sizes.
Next, the 3rd layer and the 5th layer is both convolutional layer, the convolution kernel of 384 and 256 3 × 3 sizes has been used respectively, Then reuse the processing of ReLU activation primitive；4th layer has used the pondization of 256 3 × 3 sizes to check Feature Mapping progress maximum Pondization processing；6th layer of 256 Feature Mapping for completing front convolution are transformed into the feature vector that length is 4096；7th layer right 6th layer of result carries out primary full attended operation again, is similarly obtained the feature vector that length is 4096.
Optionally, the method also includes: joined in the model lose data (dropout), local acknowledgement is returned One changes (Local Response Normalization, abbreviation LRN) and L2 canonical strategy.The principle of Dropout is with certain Probability inhibits neuron at random, and neuron a part of in this way is exactly dormant state in the training stage, does not participate in propagated forward and anti- To regeneration processes, the connectivity between network is reduced, to improve the robustness of model.The effect of LRN is to local mind Competition mechanism is established through member, so that it enhances strong neuron effect is responded, and inhibits to feed back lesser neuron.L2 model Number is the square root of institute's directed quantity quadratic sum, and L2 norm is added after the objective function of CNN can make during its minimum It obtains L2 norm and also wants minimum, and parameter therein is the weight matrix of obtained convolution kernel when training, and is being minimized in this way Then weight matrix can also be approached 0 in journey, and the parameter of model is finally made to be closer to 0, so that the generalization ability of model It is stronger.
Using the alternative embodiment, it joined dropout, LRU and L2 canonical strategy in the model, be compared to and do not add The traditional convolutional neural networks for adding these strategies, when training reached for 2000 generation, traditional convolution neural network model is had already appeared Over-fitting situation, and the model proposed by the present invention reaches highest accuracy rate, demonstrates the model with good extensive energy Power.
Optionally, the method also includes: select Natural Gamma-ray Logging Curves form as feature, by the original number of acquisition According to being changed into image format.Since there are human errors on form is demarcated, the form chosen from log data unambiguously is special Sign, artificial interception and calibration, as shown in figure 5, being from left to right respectively box-shaped, bell, infundibulate and tooth form.By to image into Row becomes standardization, addition noise, the methods of rotates and turn gray scale, carries out enhancing to data and handles with expansion, final foundation includes case The training and test data set of shape, bell, infundibulate and tooth form form of logs.
For example, morphological feature unambiguously is chosen from the log data for amounting to ten thousand metres, artificial interception and calibration, such as Shown in Fig. 5, from left to right respectively box-shaped, bell, infundibulate and tooth form.By carrying out becoming standardization, addition noise, rotation to image The methods of turn and turn gray scale, enhancing is carried out to data and is handled with expansion, final establish includes box-shaped, bell, infundibulate and tooth form The training and test data set of form of logs amount to more than 4000 pictures.
Gamma ray log response is mainly the naturally radioactive in stratum, and this radioactivity is very normal in clay mineral See, so the high mud stone of shale content and the high sandstone of quartziferous content can be distinguished, gamma ray curve can also react vertical The relative amount of sandstone and mud stone into sequence.Natural Gamma-ray Logging Curves are able to reflect depositional environment, sedimentary rhythm and deposition The geologic features such as rate, have it is good indicative, can be used in SEDIMENTARY FACIES ANALYSIS, i.e., reacted using its form of logs Type of sedimentary facies.
Since under different depositional environments, the difference of hydrodynamic condition and material resource situation will cause form and the spy of deposition The difference of sign, what is showed on Natural Gamma-ray Logging Curves is the difference of its form.Natural Gamma-ray Logging Curves form Mainly there are box-shaped, bell, infundibulate, tooth form and complex morphological, complex morphological is the combination based on preceding several forms.Box-shaped can Reflect the deposition of the rapid accumulation or ambient stable under hydrodynamic condition is stablized；The bell flow energy that is able to reflect gradually subtracts The supply of weak or material resource is fewer and fewer；Infundibulate with it is bell exactly the opposite, reflection is that hydrodynamic force becomes strong or material resource supply is got over Come more.
When identifying practical logging data, need for one section of depth sequence datas evidence to be split, and be converted to figure As form transfers to model to identify.Since the numerical value of gamma ray log data is between 0-200, the method also includes: It is first threshold value with 120, finds out the extreme point that GR numerical value is higher than 120, data is divided primary；Then it is threshold value with 100, finds out It is greater than 100 extreme value between previous segmentation section, secondary splitting is carried out to data；Be again threshold value with 85, find out previous segmentation section it Between be greater than 85 extreme value, data are divided three times；These data segments are finally converted into 2-D gray image form.
GR well logging is to the radioactive detection in stratum, and curvilinear motion is the deposition ring being able to reflect on different time scales Border variation, to indicate that well logging mutually changes with sedimentary facies.After decomposing, the geological Significance of GR curve is changed, curve Itself no longer reflects the geologic features closely related with stratum radioactivity such as shale content, still, by above-mentioned multiple dimensioned Decompose, the interface of the vertical variations trend of GR curve and variation can significantly more efficient instruction deposit level, to reach stroke Divide the purpose of different level stratigraphic units.
Optionally, the method also includes: select Daubechies wavelet basis to log data carry out different scale it is small Wave conversion, so that log data has different minutias, it is common Noise reducing of data wavelet function.Daubechies wavelet basis It is abbreviated as dbN, N indicates the order of small echo, and N is bigger, and expression wavelet filter is longer.Db small echo also can be carried out multilayer decomposition, decompose The number of plies is bigger, and expression signal decomposition number is more.By different rank and the small wave decomposition discovery of different number of plies db, 6 rank db are selected 4 layers of (abbreviation db6) small echo decompose and 4 layers of small echo of 4 rank db (abbreviation db4) decompose can effectively to log data carry out noise reduction and Filtering processing, while the subject form and growth trend of Natural Gamma-ray Logging Curves are not destroyed.As shown in fig. 6, by left-to-right point Be not original logging curve, by 4 layers of 6 rank db small echo decompose after logging curve and by 4 layers of 4 rank db small echo decompose after survey Well curve graph.It is decomposed to initial data denoising and smoothing processing, further it can be seen from this figure that carrying out 4 layers using 6 rank small echos By scaling up, in the case where using 4 rank db small echos to carry out 4 layers of decomposition so that log general form is constant, by same depth The details filtering in range in GR log is spent, only retains main trend, to divide reasonable recognition unit.Recycle institute Model is stated to identify the well logging phase element divided by different scale wavelet transformation.
Using the alternative embodiment, one-dimensional data is converted into two dimensional image and is identified, so small to data progress The trend that data are smoothed and keep original subject form and curve increase and decrease mainly is considered when wave conversion, and is selected The Daubechies small echo of suitable order and Decomposition order can keep the overall trend of curve, and will not be in model to big ruler Identification causes adverse effect when degree well logging phase；Daubechies small echo supports change of scale, can according to real data feature and The free change of scale of accuracy of identification；Daubechies small echo has good regularity, enables to signal reconstruction process more It is smooth.
In order to verify the validity of the well-log facies recognition model, well logging phase data collection on tested, and with biography The sorting algorithm of system: backpropagation (Back Propagation, abbreviation BP) neural network, support vector machine (Support Vector Machine, abbreviation SVM) and traditional convolutional neural networks (not adding dropout, LRU and L2 canonical) compared Compared with as a result such as table 1, it was demonstrated that the constructed well logging phase disaggregated model based on convolutional neural networks of the invention is mutually classified in well logging Middle performance is more preferable.
The different category of model performances of table 1
The multiple dimensioned well-log facies recognition method based on convolutional neural networks that the invention proposes a kind of, using L2 norm, Dropout and LRN effectively limits the complexity of model, improves model generalization ability, obtains compared to other calculations The category of model accuracy rate of the 97.87% of method, so as to more accurately automatic identification form of logs.
What it is due to method selection of the invention is that the well logging of two-dimensional image data is mutually used as mode to classify, unrelated specific Numerical value combines the depositional environment background to be analyzed area method proposed by the invention can be used for electrofacies analysis, So having generalization ability more preferably than classical way.
Method of the invention is divided using the well logging phase element that Daubechies small echo carries out different scale, knows each other well logging Not working becomes more convenient and efficient.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.

Claims (6)

1. a kind of multiple dimensioned well-log facies recognition method based on convolutional neural networks, which comprises the following steps:
Step (a) acquires initial data；
Step (b) establishes training dataset；
Step (c) constructs the well logging phase disaggregated model based on convolutional neural networks, identifies to data to be identified, the mould Type includes 8 layer networks, wherein the 1st layer, the 3rd layer and the 5th layer is convolutional layer, for extracting feature；2nd layer and the 4th layer is pond Layer, for reducing model complexity；6th layer and the 7th layer be full articulamentum, for by two dimensional character be changed into one-dimensional characteristic to Amount；8th layer is output layer, for classifying.
2. the multiple dimensioned well-log facies recognition method based on convolutional neural networks as described in claim 1, which is characterized in that 1 layer carries out convolution operation to input picture using convolution kernel, is then handled using ReLU activation primitive each click-through line activating, ReLU function formula is as follows:
F (x)=max (0, x)
Wherein, x is the point value, and f (x) is the new numerical value obtained by activation primitive.
3. the multiple dimensioned well-log facies recognition method based on convolutional neural networks as described in claim 1, which is characterized in that also wrap It includes: joined in the model and lose data, local acknowledgement's normalization and L2 canonical strategy.
4. the multiple dimensioned well-log facies recognition method based on convolutional neural networks as described in claim 1, which is characterized in that also wrap It includes: selecting Natural Gamma-ray Logging Curves form as feature, the initial data of acquisition is changed into image format.
5. the multiple dimensioned well-log facies recognition method based on convolutional neural networks as described in claim 1, which is characterized in that also wrap It includes: first with 120 for threshold value, finding out the extreme point that GR numerical value is higher than 120, data are divided primary；It then is threshold value with 100, The extreme value between previous segmentation section greater than 100 is found out, secondary splitting is carried out to data；It is again threshold value with 85, finds out previous segmentation It is greater than 85 extreme value between section, data is divided three times；These data segments are finally converted into 2-D gray image form.
6. the multiple dimensioned well-log facies recognition method based on convolutional neural networks as described in claim 1, which is characterized in that also wrap It includes: Daubechies wavelet basis being selected to carry out the wavelet transformation of different scale to log data.
CN201811412931.7A 2018-11-23 2018-11-23 A kind of multiple dimensioned well-log facies recognition method based on convolutional neural networks Pending CN109670533A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811412931.7A CN109670533A (en) 2018-11-23 2018-11-23 A kind of multiple dimensioned well-log facies recognition method based on convolutional neural networks

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811412931.7A CN109670533A (en) 2018-11-23 2018-11-23 A kind of multiple dimensioned well-log facies recognition method based on convolutional neural networks

Publications (1)

Publication Number Publication Date
CN109670533A true CN109670533A (en) 2019-04-23

Family

ID=66142193

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811412931.7A Pending CN109670533A (en) 2018-11-23 2018-11-23 A kind of multiple dimensioned well-log facies recognition method based on convolutional neural networks

Country Status (1)

Country Link
CN (1) CN109670533A (en)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110515123A (en) * 2019-08-22 2019-11-29 中国海洋大学 Convolutional neural networks well-shooting joint inversion method based on small-sample learning
CN111523645A (en) * 2020-04-16 2020-08-11 北京航天自动控制研究所 Convolutional neural network design method for improving small-scale target detection and identification performance
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
CN112016477A (en) * 2020-08-31 2020-12-01 电子科技大学 Logging deposition microphase identification method based on deep learning
CN112733449A (en) * 2021-01-11 2021-04-30 中国海洋大学 CNN well-seismic joint inversion method, CNN well-seismic joint inversion system, CNN well-seismic joint inversion storage medium, CNN well-seismic joint inversion equipment and CNN well-seismic joint inversion application
CN112949719A (en) * 2021-03-03 2021-06-11 中国石油大学(华东) Well testing interpretation proxy model generation method based on GAN
CN113343574A (en) * 2021-06-21 2021-09-03 成都理工大学 Mishrif group lithology logging identification method based on neural network
CN116070089A (en) * 2023-02-21 2023-05-05 北京金阳普泰石油技术股份有限公司 Stratum division method and device based on ResNet regression model and computer equipment

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110515123A (en) * 2019-08-22 2019-11-29 中国海洋大学 Convolutional neural networks well-shooting joint inversion method based on small-sample learning
CN111523645A (en) * 2020-04-16 2020-08-11 北京航天自动控制研究所 Convolutional neural network design method for improving small-scale target detection and identification performance
CN111523645B (en) * 2020-04-16 2023-04-18 北京航天自动控制研究所 Convolutional neural network design method for improving small-scale target detection and identification performance
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
CN112016477A (en) * 2020-08-31 2020-12-01 电子科技大学 Logging deposition microphase identification method based on deep learning
CN112733449A (en) * 2021-01-11 2021-04-30 中国海洋大学 CNN well-seismic joint inversion method, CNN well-seismic joint inversion system, CNN well-seismic joint inversion storage medium, CNN well-seismic joint inversion equipment and CNN well-seismic joint inversion application
CN112733449B (en) * 2021-01-11 2022-12-02 中国海洋大学 CNN well-seismic joint inversion method, CNN well-seismic joint inversion system, CNN well-seismic joint inversion storage medium, CNN well-seismic joint inversion equipment and CNN well-seismic joint inversion application
CN112949719A (en) * 2021-03-03 2021-06-11 中国石油大学(华东) Well testing interpretation proxy model generation method based on GAN
CN113343574A (en) * 2021-06-21 2021-09-03 成都理工大学 Mishrif group lithology logging identification method based on neural network
CN116070089A (en) * 2023-02-21 2023-05-05 北京金阳普泰石油技术股份有限公司 Stratum division method and device based on ResNet regression model and computer equipment

Similar Documents

Publication Publication Date Title
CN109670533A (en) A kind of multiple dimensioned well-log facies recognition method based on convolutional neural networks
Zhu et al. Intelligent logging lithological interpretation with convolution neural networks
US11320551B2 (en) Training machine learning systems for seismic interpretation
An et al. Deep convolutional neural network for automatic fault recognition from 3D seismic datasets
Ji et al. Seabed sediment classification using multibeam backscatter data based on the selecting optimal random forest model
CN110609320B (en) Pre-stack seismic reflection pattern recognition method based on multi-scale feature fusion
CN109709603A (en) Seismic horizon identification and method for tracing, system
CN107688201B (en) RBM-based seismic prestack signal clustering method
CN106372402A (en) Parallelization method of convolutional neural networks in fuzzy region under big-data environment
US11403495B2 (en) Using synthetic data sets to train a neural network for three-dimensional seismic fault segmentation
US11169287B2 (en) Method and system for automated velocity model updating using machine learning
Jiang et al. Lithology identification from well-log curves via neural networks with additional geologic constraint
CN116168295B (en) Lithology remote sensing intelligent interpretation model establishment method and interpretation method
CN112731522B (en) Intelligent recognition method, device and equipment for seismic stratum and storage medium
US20220206176A1 (en) System and method for identifying subsurface structures
CN113902861A (en) Three-dimensional geological modeling method based on machine learning
Wang et al. Seismic velocity inversion transformer
CN116824585A (en) Aviation laser point cloud semantic segmentation method and device based on multistage context feature fusion network
Wang et al. Seismic stratum segmentation using an encoder–decoder convolutional neural network
WO2021258096A1 (en) Classifying geologic features in seismic data through image analysis by neural network
CN115494560A (en) Novel modeling method based on combination of CT (computed tomography) and drilling data
Ren et al. Valuable data extraction for resistivity imaging logging interpretation
CN106997060A (en) A kind of seismic multi-attribute fusion method based on Shearlet fastICA
Lanier et al. Seafloor habitat mapping on the Oregon continental margin: A spatially nested GIS approach to mapping scale, mapping methods, and accuracy quantification
CN102360497B (en) SAR (synthetic aperture radar) image segmentation method based on parallel immune clone clustering

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20190423

WD01 Invention patent application deemed withdrawn after publication