CN106372402A - Parallelization method of convolutional neural networks in fuzzy region under big-data environment - Google Patents

Abstract

The invention discloses a parallelization method of convolutional neural networks in a fuzzy region under a big-data environment. The parallelization method comprises the following steps: firstly, constructing the convolutional neural networks in the fuzzy region, putting a given target assumption region and object identification into the same network, carrying out convolutional calculation, and updating the weight of the whole network in a training process; and secondly, dividing an input log data set into a plurality of small data sets, introducing multiple workflows to pass through the convolutional neural networks in the fuzzy region in parallel for convolution and pooling, and independently training each small data set by virtue of gradient descent. By virtue of the parallelization method, a network structure and parameters are optimized, and relatively good analysis performance and precision are realized; furthermore, the number of FR-CNN obfuscation layers is adjusted aiming at different log data sets, so that the extracted features can well reflect the characters of oil-gas reservoirs, and the fuzzification problem of the log data can be solved; and the parallel training and execution of FR-CNN are carried out by virtue of multiple GPUs, so that the efficiency of the FR-CNN is improved.

Description

The parallel method of fuzzy region convolutional neural networks under a kind of big data environment

Technical field

The present invention relates to petroleum well logging technology field, more particularly to big data well logging field.

Background technology

Well logging information is with reflection and governing factor that deposition is formation rock physical property, therefore well-log information all the time By the important information source as basis in oil and gas reservoir sedimentology research, well logging phase is then well logging information and Reservoir Sedimentological Bridge between feature.For most Oil/gas Well, well-log information is only synthesis letter covering full well section stratum Breath source, therefore well-log facies recognition analysis method is always as a most important research in oil exploration and exploitation geological research Means.

However, well logging information has the characteristics that ambiguity, there is multi-solution and the ambiguity of geological Significance.Therefore, log well The identification of phase and analysis must be set up dividing with log parameter relation (log response) synthesis depth in existing deposition characteristicses in a large number On analysis basis, the result of referring also to outcrop, core log and earthquake analysis simultaneously, choose and be suitable for building of geology characteristic Mould method, just enables accurately identifying of well logging phase.

Further, since lacking phase automatic identifying method and the technology of effectively logging well, current well-log facies recognition is mainly logical Cross geological work personnel artificial cognition realize, and due to personnel's experience difference, subjective differences, log data System level gray correlation The factor such as different, the data volume faced by geological personnel is big, workload is heavy.Moreover, the experience difference of geological personnel, subjective because Element, the factor such as systematical difference of different times difference instrument log data are so that the traditional big discounting of well-log facies recognition accuracy Button.

Big data is analyzed, the advanced technology such as deep learning is applied to oil-gas geology research is that to solve current oil industry big The exploration of data analysiss resources idle and trial.In recent years, petroleum industry establishes substantial amounts of cloud data center, but utilization rate is not Height, resource is by serious waste.One of major reason is just a lack of big data processing platform and corresponding big data technology To make full use of these calculating, storage resources.

Set up the urgent needss that efficient, accurate well-log facies recognition method is present oil-gas geology research.

Content of the invention

For solving the deficiencies in the prior art, the present invention proposes fuzzy region convolutional neural networks under a kind of big data environment Parallel method.

The technical scheme is that and be achieved in that:

The parallel method of fuzzy region convolutional neural networks under a kind of big data environment, first, builds fuzzy region volume Long-pending neutral net, will provide goal hypothesis region and target recognition is put in same network, shared convolutional calculation, a training Process updates the weight of whole network；

Next, the log data collection of input is divided into some small data set, multiple workflow parallelizations are through fuzzy Region convolutional neural networks carry out convolution and pondization operation, and each small data set is trained individually with gradient decline；Training After the completion of, result is exported waiting list, after the completion of a wheel training, reads output queue, carry out the synchronization of shared weight Update operation, after the completion of renewal, carry out next round training；In each wheel training, the meter of the small data set that each is split Calculate, asynchronous all on distributed basis carry out, often calculate Grad, be just appended in the middle of list come, when all of little Data set all calculates after finishing, the weight of synchronized update fuzzy region convolutional neural networks and bias, then carries out next round Training；In terms of parallelization identification, log data is collected by spout, then that data distribution is parallel in each bolt node Carry out well-log facies recognition, recognition result is input in next bolt node each bolt node, statistics object therein letter Breath；

The step that each small data set carries out convolution and pondization operation through fuzzy region convolutional neural networks, concrete bag Include: convolutional layer and the interaction of pond layer, carry out fuzzy operation in convolutional layer and pond layer, from the of fuzzy region convolutional neural networks One layer starts, and is gradually increased the number of plies of obfuscation, adjusts the obfuscation number of plies, fuzzy region convolutional Neural for different data sets Last layer of network obtains characteristic vector, and this feature vector passes through a sliding window by Feature Mapping to a low-dimensional vector In, then input the feature into two full articulamentums, a full articulamentum is used for positioning, and another full articulamentum is used for classifying.

Alternatively, described convolutional layer formula is expressed as:

$v_{ij}^{xy}=f({\stackrel{\&overbar;}{b}}_{ij}+{\σ}_m{\σ}_{p=0}^{p_i-1}{\σ}_{q=0}^{q_i-1}{\stackrel{\&overbar;}{w}}_{ijm}^{pq}{v}_{(i-1)m}^{(x+p)(y+q)})$

Pond layer formula is expressed as:

$x_j=f\left({\stackrel{\&overbar;}{\mathrm{\β}}}_{ij}down\right(x_{i-1j})+{\stackrel{\&overbar;}{b}}_{ij})$

Wherein, biasAnd weightIt is fuzzy number, used here as symmetrical triangular fuzzy numbers,For The vector of fuzzy number composition, j-th fuzzy numberMembership function be:

${\stackrel{\&overbar;}{w}}_j\left(w\right)=\max \left\{1-\frac{|w-w_j|}{{\hat{w}}_j},0\right\}.$

Alternatively, in the training process of fuzzy region convolutional neural networks, one associated losses function of definition:

$l\left(\left\{p_i\right\},\left\{t_i\right\}\right)=\frac{1}{n_{cls}}{\σ}_i{l}_{cls}(p_i,p_i^{*})+\mathrm{\λ}\frac{1}{n_{reg}}{\σ}_i{p}_i^{*}{l}_{reg}(t_i,t_i^{*})$

Wherein, p_iIt is the prediction probability that this sample is form of logs,It is the label of sample, if corresponding survey Well tracing pattern,For 1, otherwiseFor 0, n_clsIt is two sorted logic losses；t_iIt is four parameter compositions of prediction object boundary Vector,For the vector of tab area parameter composition, they are respectively as follows:

t_x=(x-x_a)/w_at_h=(y-y_a)/h_a

t_w=log (w/w_a) t_h=log (h/h_a)

$\begin{array}{cc}{t}_w^{*}=(w^{*}/w_a)/w_a& t_y^{*}=(y*-y_a)/h_a\end{array}$

$\begin{array}{cc}{t}_w^{*}=log(w^{*}/w_a)& t_h^{*}=log(h*/h_a)\end{array}$

Wherein, x, y, w and h represent centre coordinate, width and the length of object, x, x respectively_a, x^*Represent Target area respectively Domain, anchor region and tab area, return lossR is smooth loss function

${\mathrm{smooth}}_{l_1}\left(x\right)=\left\{\begin{array}{cc}0.5x^2& if\left|x\right|<1\\ \left|x\right|-0.5& otherwise\end{array}\right.$

Represent that only working as anchor region is positive sampleWhen, just calculate and return loss, otherwiseDo not calculate, Normalized parameter n_clsAnd n_regRepresent the length of low-dimensional vector from maps feature vectors and the quantity of anchor region respectively.

Alternatively, carry out the standardization of log data first, initial data be converted into index without dimension test and appraisal value, Each index test and appraisal value is all in same number of levels, then carries out comprehensive test analysis.

Alternatively, the standardization carrying out log data is using following normalization method:

Sx=(x-m)/s, x ∈ gr, ac, den, cnl, sdn ... }

Wherein, x represents the data of every log, and sx represents the borehole log data after standardization, and m is corresponding well logging The average of curve data, s is the standard deviation of every borehole log data.

The invention has the beneficial effects as follows:

(1) feature being obscured according to data in well logging big data, incorporates fuzzy theory, proposes fuzzy region convolutional Neural net Network fr-cnn, and progressive fuzzy method is proposed, from the beginning of the ground floor of fuzzy region convolutional neural networks, it is gradually increased fuzzy The number of plies changed, thus optimizing network structure and parameter, realizes more preferable analytical performance and precision；

(2) it is directed to the number of plies that different log data collection adjust fr-cnn obfuscation, so that the feature of extraction is preferably reflected The characteristic of oil and gas reservoir itself, can solve log data fuzzy problem；

(3) present invention carries out parallel training and the execution of fr-cnn using many gpu, to improve the efficiency of fr-cnn.

Brief description

In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing Have technology description in required use accompanying drawing be briefly described it should be apparent that, drawings in the following description be only this Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, acceptable Other accompanying drawings are obtained according to these accompanying drawings.

Fig. 1 is the structural representation of fuzzy region convolutional neural networks of the present invention；

Fig. 2 is symmetrical triangular fuzzy numbers coordinate schematic diagram；

Fig. 3 is the schematic diagram that fuzzy region convolutional neural networks parallelization of the present invention processes real time data.

Specific embodiment

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete Site preparation description is it is clear that described embodiment is only a part of embodiment of the present invention, rather than whole embodiments.It is based on Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of not making creative work Embodiment, broadly falls into the scope of protection of the invention.

Log data has the characteristics that ambiguity, is many reason cause this ambiguity, including noise, differs The data space pollution of the log data that cause property, imperfection etc. cause, also includes different times, different instrument well logging brings Systematic data difference, the log data ambiguity that these problems are brought all constrains accurately identifying of well logging phase.

The present invention proposes a kind of parallel method of fuzzy region convolutional neural networks under big data environment, to well logging number According to the data space constructing various dimensions, fuzzy theory is merged with deep learning network r-cnn, propose to solve fuzzy data feelings The recognition methodss of phase of logging well under condition, the feature being obscured according to data in well logging big data, incorporate fuzzy theory, propose fuzzy region Convolutional neural networks fr-cnn (fuzzy r-cnn), it is further proposed that progressive blur method, opens from convolutional neural networks ground floor Begin, be gradually increased the number of plies of obfuscation, optimize network structure and parameter, finally set up theory and the method for fr-cnn, realize more Good analytical performance and precision, meanwhile, the present invention carries out parallel training and the execution of fr-cnn using many gpu, to improve fr- The efficiency of cnn.

Design the emphasis that suitable fuzzy region convolutional neural networks are the present invention, below to fuzzy region convolution of the present invention The structure of neutral net is described in detail.

Fuzzy region convolutional neural networks fr-cnn builds on the basis of deep learning network r-cnn, as Fig. 1 institute Show, fr-cnn will provide goal hypothesis region and target recognition is put in same network, shared convolutional calculation, it is to avoid complicated Calculation procedure, it is only necessary to training process just can update the weight of whole network, also accelerates detection speed simultaneously, reaches fast The purpose that speed is processed.

In Fig. 1, log data, through fuzzy region convolutional neural networks, carries out convolution and pondization operation.Fuzzy region is rolled up The core of long-pending neural metwork training is convolutional layer and the interaction of pond layer, therefore carries out fuzzy behaviour in convolutional layer and pond layer Make.In order to avoid the fuzzy information loss excessively leading to is excessive, and extract feature in view of fuzzy region convolutional neural networks The degree that becomes more meticulous successively reduce, here change the obfuscation to each layer for the traditional fuzzy neutral net, the present invention propose progressive Fuzzy method, that is, from the beginning of the ground floor of fuzzy region convolutional neural networks, is gradually increased the number of plies of obfuscation, for difference Data set adjustment the obfuscation number of plies, make the feature of extraction preferably reflect the characteristic of log, thus obtaining best identified As a result, and improve recognition efficiency.

Last layer of fuzzy region convolutional neural networks obtains characteristic vector, and this feature vector passes through a little slip Feature Mapping in a low-dimensional vector, is then input the feature into two full articulamentums, a full articulamentum is used for by window Positioning, another full articulamentum is used for classifying.Provide several goal hypothesis regions simultaneously at each sliding window, can be referred to as For anchor region, this region, centered on sliding window, has different transverse and longitudinal ratios and scaling.

The convolutional layer formula of convolutional neural networks r-cnn can be expressed as:

$v_{ij}^{xy}=f(b_{ij}+{\&sigma;}_m{\&sigma;}_{p=0}^{p_i-1}{\&sigma;}_{q=0}^{q_i-1}{w}_{ijm}^{pq}{v}_{(i-1)m}^{(x+p)(y+q)})---\left(1\right)$

Wherein,Represent is the value at (x, y) position of j-th characteristic vector of i-th layer of neuron,Table ShowIt is connected to the convolution kernel of the m-th characteristic vector weights on position (p, q).p_iAnd q_iRepresent the height of convolution kernel respectively And width, b_ijFor bias term, the activation primitive of f (x) expression neuron.

R-cnn pond layer formula is expressed as:

x_ij=f (β_ijdown(x_i-1j)+b_ij) (2)

Down (.) represents a down-sampling function, and typical operation is usually all of the different n*n blocks to input data Information is sued for peace, and such output data all reduces n times in two dimensions, and each output map corresponds to one and belongs to certainly Oneself property taken advantage of biases β and additivity and biases b.

The input of convolutional neural networks and calculating process are all real numbers, the result obtaining all being to determine property, and for number Situation about obscuring according to data such as disappearances, introduces fuzzy theory, improved formula in the fuzzy region convolutional neural networks of the present invention As follows:

Convolutional layer formula is expressed as:

$v_{ij}^{xy}=f({\stackrel{\&overbar;}{b}}_{ij}+{\&sigma;}_m{\&sigma;}_{p=0}^{p_i-1}{\&sigma;}_{q=0}^{q_i-1}{\stackrel{\&overbar;}{w}}_{ijm}^{pq}{v}_{(i-1)m}^{(x+p)(y+q)})---\left(3\right)$

Pond layer formula is expressed as:

$x_j=f\left({\stackrel{\&overbar;}{\mathrm{\&beta;}}}_{ij}down\right(x_{i-1j})+{\stackrel{\&overbar;}{b}}_{ij})---\left(4\right)$

Wherein biasAnd weightIt is fuzzy number, used here as symmetrical triangular fuzzy numbers,For mould The vector that paste array becomes, j-th fuzzy numberMembership function be

${\stackrel{\&overbar;}{w}}_j\left(w\right)=\max \left\{1-\frac{|w-w_j|}{{\hat{w}}_j},0\right\}---\left(5\right)$

As shown in Fig. 2 w_jIt is the symmetrical centre of fuzzy number,It is half length of fuzzy number,Represent the degree of membership at w.

In the training process of fuzzy region convolutional neural networks, one associated losses function of definition:

$l\left(\left\{p_i\right\},\left\{t_i\right\}\right)=\frac{1}{n_{cls}}{\&sigma;}_i{l}_{cls}(p_i,p_i^{*})+\mathrm{\&lambda;}\frac{1}{n_{reg}}{\&sigma;}_i{p}_i^{*}{l}_{reg}(t_i,t_i^{*})---\left(6\right)$

Wherein p_iIt is the prediction probability that this sample is form of logs,It is the label of sample, if corresponding survey Well tracing pattern,For 1, otherwiseFor 0, n_clsIt is the loss of two classification (0 or 1) logic.

t_iIt is the vector of four parameter compositions of prediction object boundary,For the vector of tab area parameter composition, they divide It is not:

t_x=(x-x_a)/w_at_h=(y-y_a)/h_a(7)

t_w=log (w/w_a)t_h=log (h/h_a)

$\begin{array}{cc}{t}_x^{*}=(x^{*}-x_a)/w_a& t_y^{*}=(y^{*}-y_a)/h_a\end{array}$

$\begin{array}{cc}{t}_w^{*}=log(w^{*}/w_a)& t_h^{*}=log(h*/h_a)\end{array}$

Wherein x, y, w and h represent centre coordinate, width and the length of object, x, x respectively_a, x^*Represent estimation range respectively, Anchor region and tab area (y, w, h are in the same manner).Return lossR is smooth loss function

${\mathrm{smooth}}_{l_1}\left(x\right)=\left\{\begin{array}{cc}0.5x^2& if\left|x\right|<1\\ \left|x\right|-0.5& otherwise\end{array}\right.---\left(8\right)$

Represent only when anchor region is for positive sampleJust calculate and return loss, otherwiseDisregard Calculate.Normalized parameter n_clsAnd n_regRepresent the length of low-dimensional vector from maps feature vectors and the quantity of anchor region respectively.

Different data can be produced using different well logging means.As using natural gamma (gr), compensation sound wave (ac), mended Repay density (den), compensated neutron (cnl) and neutron apparent porosity poor from density apparent porosity (sdn) etc. and there are different dimensions, There is no between data comparability, therefore, the present invention needs to carry out the standardization of log data first, and initial data is all changed For index without dimension test and appraisal value, that is, each index test and appraisal value is all in same number of levels, then carries out comprehensive test analysis.

Using following normalization method:

Sx=(x-m)/s, x ∈ gr, ac, den, cnl, sdn ... }

Wherein, x represents the data of every log, and sx represents the borehole log data after standardization；M is corresponding well logging The average of curve data, s is the standard deviation of every borehole log data.

The present invention is demarcated from existing log data, sets up the training dataset of fr-cnn, on this basis, by Information disclosed in different logging methods is not quite similar, so selecting combining of different log datas defeated as fr-cnn Enter, so that it is determined that the optimum log data combination of fr-cnn, and optimize the network parameter carrying out during well-log facies recognition of fr-cnn And structure.

Fr-cnn two classes full articulamentums more than traditional convolutional neural networks, more than the operation such as the calculating of area coordinate, These operation amounts of calculation are all very big.In fuzzy neural network, fuzzy operation is present in each layer of network that is to say, that network Deeper increased amount of calculation is more, and this just makes originally to need the network of heavy calculating to seem heavy.The increase of amount of calculation is led The training time causing network increases substantially, and extends the cycle of network model's renewal, weakens the motility of system, when detecting simultaneously Between also can lengthen.

The present invention improves fr-cnn training and operational efficiency by parallelization, first the log data collection of input is divided into Some small data set, multiple workflows are run simultaneously, and each section is trained individually with gradient decline.After the completion of training, Result is exported waiting list, after the completion of a wheel training, reads output queue, carry out the synchronized update behaviour of shared weight Make.After the completion of renewal, carry out next round training.

Each wheel training in, the calculating of the small data set that each is split, be all on distributed basis asynchronous enter Row, often calculate Grad, be just appended to and come in the middle of list, after all of small data set all calculates and finishes, synchronized update The weight of network and bias, then carry out next round training.

As shown in figure 3, in terms of parallelization identification, the solution taken is: log data is collected by spout, then Data distribution is carried out well-log facies recognition parallel in each bolt node, recognition result is input to next by each bolt node In individual bolt node, count object information therein.

The feature that the present invention obscures according to data in well logging big data, incorporates fuzzy theory, proposes fuzzy region convolution god Through network fr-cnn, and propose progressive fuzzy method, from the beginning of the ground floor of fuzzy region convolutional neural networks, be gradually increased The number of plies of obfuscation, thus optimizing network structure and parameter, realizes more preferable analytical performance and precision；And, the present invention is directed to Different log data collection adjusts the number of plies of fr-cnn obfuscation, makes the feature of extraction preferably reflect the spy of oil and gas reservoir itself Property, log data fuzzy problem can be solved；For operating computationally intensive problem, the present invention carries out fr- using many gpu The parallel training of cnn and execution, to improve the efficiency of fr-cnn.

The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all essences in the present invention Within god and principle, any modification, equivalent substitution and improvement made etc., should be included within the scope of the present invention.

Claims (5)

1. under a kind of big data environment fuzzy region convolutional neural networks parallel method it is characterised in that

First, build fuzzy region convolutional neural networks, goal hypothesis region will be given and target recognition puts into same network In, shared convolutional calculation, a training process updates the weight of whole network；

Next, the log data collection of input is divided into some small data set, multiple workflow parallelizations are through fuzzy region Convolutional neural networks carry out convolution and pondization operation, and each small data set is trained individually with gradient decline；Training completes Afterwards, result is exported waiting list, after the completion of a wheel training, read output queue, carry out the synchronized update of shared weight Operation, after the completion of renewal, carries out next round training；In each wheel training, the calculating of the small data set that each is split, all Asynchronous on distributed basis carry out, often calculate Grad, be just appended in the middle of list come, when all of small data set All calculate after finishing, the weight of synchronized update fuzzy region convolutional neural networks and bias, then carry out next round training；? Parallelization identification aspect, collects log data by spout, then data distribution is logged well parallel in each bolt node Mutually identify, recognition result is input in next bolt node each bolt node, count object information therein；

The step that each small data set carries out convolution and pondization operation through fuzzy region convolutional neural networks, specifically includes: Convolutional layer and the interaction of pond layer, carry out fuzzy operation in convolutional layer and pond layer, from the first of fuzzy region convolutional neural networks Layer starts, and is gradually increased the number of plies of obfuscation, adjusts the obfuscation number of plies, fuzzy region convolutional Neural net for different data sets Last layer of network obtains characteristic vector, and this feature vector passes through a sliding window by Feature Mapping to a low-dimensional vector In, then input the feature into two full articulamentums, a full articulamentum is used for positioning, and another full articulamentum is used for classifying.

2. as claimed in claim 1 under a kind of big data environment fuzzy region convolutional neural networks parallel method, it is special Levy and be, described convolutional layer formula is expressed as:

$v_{ij}^{xy}=f({\stackrel{\&overbar;}{b}}_{ij}+{\mathrm{\&sigma;}}_m{\mathrm{\&sigma;}}_{p=0}^{p_i-1}{\mathrm{\&sigma;}}_{q=0}^{q_i-1}{\stackrel{\&overbar;}{w}}_{ijm}^{pq}{v}_{(i-1)m}^{(x+p)(y+q)})$

Pond layer formula is expressed as:

$x_j=f\left({\stackrel{\&overbar;}{\mathrm{\&beta;}}}_{ij}down\right(x_{i-1j})+{\stackrel{\&overbar;}{b}}_{ij})$

Wherein, biasAnd weightIt is fuzzy number, used here as symmetrical triangular fuzzy numbers,For fuzzy number The vector of composition, j-th fuzzy numberMembership function be:

${\stackrel{\&overbar;}{w}}_j\left(w\right)=\max \{1-\frac{\left|w-w_j\right|}{{\hat{w}}_j},0\}.$

3. as claimed in claim 2 under a kind of big data environment fuzzy region convolutional neural networks parallel method, it is special Levy and be, in the training process of fuzzy region convolutional neural networks, one associated losses function of definition:

$l\left(\right\{p_i\},\{t_i\left\}\right)=\frac{1}{n_{cls}}{\mathrm{\&sigma;}}_i{l}_{cls}(p_i,p_i^{*})+\mathrm{\&lambda;}\frac{1}{n_{reg}}{\mathrm{\&sigma;}}_i{p}_i^{*}{l}_{reg}(t_i,t_i^{*})$

Wherein, p_iIt is the prediction probability that this sample is form of logs,It is the label of sample, if corresponding song of logging well Line morphology,For 1, otherwiseFor 0, n_clsIt is two sorted logic losses；t_iBe prediction object boundary four parameters composition to Amount,For the vector of tab area parameter composition, they are respectively as follows:

t_x=(x-x_a)/w_at_h=(y-y_a)/h_a

t_w=log (w/w_a) t_h=log (h/h_a)

$t_x^{*}=(x^{*}-x_a)/w_a,t_y^{*}=(y^{*}-y_a)/h_a$

$t_w^{*}=\log (w^{*}/w_a),t_h^{*}=\log (h^{*}/h_a)$

Wherein, x, y, w and h represent centre coordinate, width and the length of object, x, x respectively_a, x^*Represent estimation range, anchor respectively Determine region and tab area, return lossR is smooth loss function

${\mathrm{smooth}}_{l_1}\left(x\right)=\left\{\begin{array}{cc}0.5x^2& if\left|x\right|<1\\ \left|x\right|-0.5& otherwise\end{array}\right.$

Represent that only working as anchor region is positive sampleWhen, just calculate and return loss, otherwiseDo not calculate, normalizing Change parameter n_clsAnd n_regRepresent the length of low-dimensional vector from maps feature vectors and the quantity of anchor region respectively.

4. under a kind of big data environment as described in any one of claims 1 to 3 fuzzy region convolutional neural networks parallelization Method, it is characterised in that carrying out the standardization of log data first, initial data is converted into index without dimension test and appraisal Value, each index test and appraisal value is all in same number of levels, then carries out comprehensive test analysis.

5. as claimed in claim 4 under a kind of big data environment fuzzy region convolutional neural networks parallel method, it is special Levy and be, the standardization carrying out log data is using following normalization method:

Sx=(x-m)/s, x ∈ gr, ac, den, cnl, sdn ... }

Wherein, x represents the data of every log, and sx represents the borehole log data after standardization, and m is corresponding log The average of data, s is the standard deviation of every borehole log data.