CN110552693A - layer interface identification method of induction logging curve based on deep neural network - Google Patents
layer interface identification method of induction logging curve based on deep neural network Download PDFInfo
- Publication number
- CN110552693A CN110552693A CN201910886031.4A CN201910886031A CN110552693A CN 110552693 A CN110552693 A CN 110552693A CN 201910886031 A CN201910886031 A CN 201910886031A CN 110552693 A CN110552693 A CN 110552693A
- Authority
- CN
- China
- Prior art keywords
- logging
- layer
- segment
- data
- processed
- 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
Classifications
-
- E—FIXED CONSTRUCTIONS
- E21—EARTH DRILLING; MINING
- E21B—EARTH DRILLING, e.g. DEEP DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS
- E21B49/00—Testing the nature of borehole walls; Formation testing; Methods or apparatus for obtaining samples of soil or well fluids, specially adapted to earth drilling or wells
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V3/00—Electric or magnetic prospecting or detecting; Measuring magnetic field characteristics of the earth, e.g. declination, deviation
- G01V3/18—Electric or magnetic prospecting or detecting; Measuring magnetic field characteristics of the earth, e.g. declination, deviation specially adapted for well-logging
- G01V3/26—Electric or magnetic prospecting or detecting; Measuring magnetic field characteristics of the earth, e.g. declination, deviation specially adapted for well-logging operating with magnetic or electric fields produced or modified either by the surrounding earth formation or by the detecting device
- G01V3/28—Electric or magnetic prospecting or detecting; Measuring magnetic field characteristics of the earth, e.g. declination, deviation specially adapted for well-logging operating with magnetic or electric fields produced or modified either by the surrounding earth formation or by the detecting device using induction coils
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V3/00—Electric or magnetic prospecting or detecting; Measuring magnetic field characteristics of the earth, e.g. declination, deviation
- G01V3/38—Processing data, e.g. for analysis, for interpretation, for correction
Abstract
The invention relates to the technical field of layer interface identification of geophysical induction logging curves, in particular to a layer interface identification method of an induction logging curve based on a deep neural network, which comprises the following steps: windowing and cutting the logging curve acquired in real time to obtain a plurality of logging segments; carrying out logarithm and normalization processing on all logging values on each logging segment to obtain logging segment data corresponding to each processed logging segment; inputting the acquired logging segment data corresponding to each processed logging segment into a pre-trained deep neural network classification model, and acquiring the type and probability corresponding to each logging segment data; and judging whether a layer interface exists between every two adjacent logging fragment data or not by comparing the corresponding types and the corresponding probabilities of every two adjacent logging fragment data.
Description
Technical Field
the invention belongs to the technical field of layer interface identification of geophysical logging and induction logging curves, and particularly relates to a layer interface identification method of an induction logging curve based on a deep neural network.
background
Geophysical logging is a method for measuring the physical properties of the formation, such as electricity, sound, radioactivity, etc., in a borehole to distinguish the properties of the formation rocks and fluids, and is an important means for exploring and developing mineral resources, such as oil and gas, metals, etc. As shown in fig. 1, the logging apparatus comprises: logging instrument 111, logging system 112, cable 113, and sheave wheel 114. The formation 116 signals are measured by a logging tool 111 disposed in the mud 115 and then transmitted through a cable to a logging system 112 for data acquisition. During the well logging operation, firstly, a well logging instrument is put down into a well bore, then, a ground system sends a well logging instruction to the downhole instrument through a cable and a remote measuring system, the well logging instrument obtains measuring signals of stratum information at different depths in a lifting or lowering mode, and the measuring signals are transmitted to the ground system through the remote measuring system and the cable. The ground system analyzes underground geological features by analyzing the measurement signals containing stratum information, obtains the distribution and trend characteristics of oil and gas resources, mineral resources and water resources in geology by analyzing and inverting the geological features, and provides geological data for the resource exploration and development process.
Induction logging is the most important measurement method in logging, and the electrical parameters of the stratum are measured by the electromagnetic induction principle. Specifically, the induction logging is implemented by transmitting an alternating current with a certain frequency through a transmitting coil, the alternating current generates an alternating electromagnetic field, a stratum medium generates a vortex under the action of the alternating electromagnetic field, the vortex forms a secondary alternating electromagnetic field, and a receiving coil generates a corresponding induced electromotive force under the action of the secondary alternating electromagnetic field, wherein the induced electromotive force has a direct relation with a geological electrical characteristic (particularly a conductive characteristic, namely a conductivity parameter).
Layer boundary information is critical to geophysical well log data interpretation. The layer interface information is the main content of geophysical logging interpretation and has a positive effect on later-stage logging data inversion work. At present, a layering method based on a logging curve carries out layer interface identification according to curve characteristics such as an inflection point position and a maximum slope change point of the curve, and the layering method is good in performance in a layer interface identification process of a simple geological model, such as boundary information of a thick layer. However, when the logging instrument is located near the thin layer, especially when the resistivity of the surrounding rock on both sides of the thin layer is very different, the effect of extracting the layer interface information by the methods is seriously reduced, and the manual correction is needed.
the traditional method for extracting layer interface information is mainly based on the characteristics of the maximum inflection point of a logging curve or the maximum change of curve inclination and the like to divide stratums, and the method has good performance in the layer interface identification process of a simple geological model, such as boundary information of a thick layer. However, when the logging instrument is located near the thin layer, especially when the resistivity difference of the surrounding rocks on the two sides of the thin layer is very large, the effect of extracting the layer interface information by the methods is seriously reduced, and the layer interface needs to be judged manually. In the prior art, the layer interface information of induction logging is identified by a neural network method, and a training set model is established for learning by the method according to the logarithmic derivative information of actually measured and simulated logging curves, but the effect is not ideal, and the layer interface is misjudged.
Disclosure of Invention
The invention aims to solve the technical defects of the existing identification method, and provides a layer interface identification method for realizing an induction logging curve based on a deep neural network model. According to the method, a deep neural network model is obtained by performing a supervised learning training process on a logging curve obtained by numerical modeling, and the recognition capability of layer interface information is realized. The method can be used for extracting the layer interface information of the induction logging response curve of the complex stratum model, and the processing method has no artificial factors and has good robustness and real-time performance.
in order to achieve the above object, the present invention provides a layer interface identification method of an induction log based on a deep neural network, which is characterized in that the method comprises:
Windowing and cutting the logging curve acquired in real time to obtain a plurality of logging segments;
carrying out logarithm and normalization processing on all logging values on each logging segment to obtain logging segment data corresponding to each processed logging segment;
Inputting the acquired logging segment data corresponding to each processed logging segment into a pre-trained deep neural network classification model, and acquiring the type and probability corresponding to each logging segment data;
and judging whether a layer interface exists between every two adjacent logging fragment data or not by comparing the corresponding types and the corresponding probabilities of every two adjacent logging fragment data.
as an improvement of the above technical solution, all the logging values on each logging segment are subjected to logarithmization and normalization processing to obtain logging segment data corresponding to each processed logging segment; the method specifically comprises the following steps:
according to the formula (1), all the logging values on each logging segment are subjected to logarithmic treatment and normalization treatment, and the logarithmic treatment and normalization treatment on all the logging values on each logging segment are completed:
Wherein the content of the first and second substances,For the ith log value R on the log sliceiLog values after logarithmic and normalization processing;
obtaining all processed logging values on each logging segment according to the formula (1), and further obtaining each processed logging segment; thereby obtaining logging segment data corresponding to each processed logging segment; and the logging segment data corresponding to each processed logging segment is all logging values on each processed logging segment.
as one improvement of the above technical solution, the obtained logging segment data corresponding to each processed logging segment is input to a pre-trained deep neural network classification model, and a category and a probability corresponding to each logging segment data are obtained; the method specifically comprises the following steps:
Inputting logging segment data corresponding to each processed logging segment into a pre-trained deep neural network classification model, and realizing dimensionality reduction processing through multilayer network layer weight and bias and loading of an activation function to obtain processed logging data; the activation function is specifically shown in formula (2):
Wherein z isi+1Is the output of the i-layer network layer; z is a radical ofiis the input of the i-layer network layer; f. ofian activation function for an i-layer network layer; withe weight matrix is the weight matrix of the i-layer network layer; biThe offset vector of the i-layer network layer;
According to the obtained processed logging data, classifying a full connection layer of the last layer of the model by using a deep neural network and an activation function soft max, and according to a formula (3), obtaining the corresponding type and probability of each logging fragment data;
Wherein z ism,kThe original output of the full connection layer of the last layer; m is the last layer, k is the kth neuron of the last layer, zm,nFor the nth neuron of the last layer, sigmanexp(zm,n) Exponential form of all neurons in the last layer, soft max (z)m,k) The original output z of the fully-connected layer being the last layerm,kOutputting after passing through the soft max function, namely, the probability of each corresponding type of the logging segment data;
If the target is marked as two types, two probabilities with different sizes can be output through a soft max function, the two probability values are compared, the larger probability value is selected as the final output and is marked as the probability corresponding to each logging segment data, the type corresponding to the larger probability is the type of the target and is marked as the type corresponding to each logging segment data.
As an improvement of the above technical solution, the deep neural network classification model specifically includes: a first input layer, a second input layer, a third input layer, a plurality of local connection layers, a multiplied fusion layer, an added fusion layer, a transition layer, and a full connection layer;
the logging value of the logging segment and the multiple characteristics of the logging segment are used as input data and are respectively input into a first input layer, a second input layer and a third input layer, dimension reduction processing of the logging segment is realized through respective local connecting layers, combination of different attributes of the logging segment is realized through multiplying the fusion layer and the added fusion layer, and finally the type and the probability of the logging segment are obtained through soft max in the transition layer and the full connecting layer.
As one improvement of the above technical solution, the method includes comparing the type and probability corresponding to each two adjacent logging segment data to determine whether a layer interface exists between the two adjacent logging segment data; the method specifically comprises the following steps:
If the type and the probability corresponding to one logging segment data are different from those corresponding to the adjacent logging segment data, a layer interface exists between the two adjacent logging data, and the two adjacent logging data are identified and marked.
Compared with the prior art, the invention has the beneficial effects that:
The layer interface recognition method obtains the knowledge of the layer interface characteristics by performing supervised learning training on the response data and the data characteristics of the induction logging instrument, and the process has no human factors, so that the reliability of layer interface recognition is improved. The layer interface identification method has good data processing efficiency and real-time performance and is suitable for field data processing.
drawings
FIG. 1 is a schematic diagram of a prior art logging tool in operation for logging;
FIG. 2 is a schematic flow chart of a method for identifying a layer interface of an induction log based on a deep neural network according to the present invention;
FIG. 3 is a block diagram of a deep neural network classification model of a method for layer interface recognition of induction logs based on a deep neural network of the present invention;
fig. 4 is a graph of the layer interface recognition effect of a chirp model (Rhigh/Rlow 100) in an embodiment of the method for recognizing the layer interface of the induction logging curve based on the deep neural network of the present invention;
FIG. 5a is a log response graph of an Oklahoma geological model in another embodiment of a method for layer interface identification of induction logs based on a deep neural network of the present invention;
Fig. 5b is a layer interface recognition effect diagram of the Oklahoma geological model in another embodiment of the induction well logging curve layer interface recognition method based on the deep neural network of the present invention.
Reference numerals:
111. logging instrument 112 and logging system
113. Cable 114 and pulley
115. mud 116, formation
Detailed Description
The invention will now be further described with reference to the accompanying drawings.
as shown in fig. 2, the present invention provides a layer interface recognition method for implementing an induction logging curve based on a deep neural network model, which includes:
Collecting an original logging curve, and preprocessing the original logging curve to obtain a plurality of logging segments; inputting the plurality of logging segments into a pre-trained deep neural network classification model to obtain the type and probability of each logging segment and the type and probability of a geological point corresponding to a logging depth; and evaluating and comparing the types and the probabilities of the geological points at two adjacent logging depths, and judging whether to mark the layer interface according to the comparison result.
the method specifically comprises the following steps:
step 1) windowing and cutting a logging curve acquired in real time to obtain a plurality of logging segments; in particular, the amount of the solvent to be used,
Determining the window width according to the longitudinal resolution of the adopted existing induction logging instrument, and windowing and cutting the logging curve acquired in real time to acquire a plurality of logging segments; the logging device comprises a logging segment, a plurality of sensing logging instruments and a plurality of logging segments, wherein each logging segment is provided with odd number of logging values, the logging depth corresponding to the central point of each logging segment is the logging depth of the logging segment, the logging segments are used for providing geological information near the measuring points for data processing of the logging values of the central points of the logging segments, and the geological information is related to the existing sensing logging instruments and geological distribution.
Step 2) carrying out logarithm and normalization processing on all logging values on each logging segment to obtain logging segment data corresponding to each processed logging segment; in particular, the amount of the solvent to be used,
and carrying out logarithm and normalization processing on the logging values of all logging segments, as shown in the following formula 1, and completing the logarithm and normalization processing on the logging values of each logging segment:
wherein the content of the first and second substances,For the ith log value R on the log sliceicarrying out logarithm and normalization processing on the logging values, recombining all the processed logging values on each logging segment to form each new logging segment, and using the new logging segment as the input of a deep neural network classification model; log (R)i) For the ith log value RiA logarithmic value of; log (2000) is a logarithmic value of 2000 ohmm; log (0.1) is the log value of 0.1 ohmm; 0.1-2000 ohmm is the dynamic measurement range of induction logging;
And (2) obtaining all the processed logging values on each logging segment according to the formula (1), and further obtaining each processed logging segment, namely each new logging segment, so as to obtain each processed logging segment data.
step 3) inputting logging segment data corresponding to each processed logging segment into a pre-trained deep neural network model, and acquiring the type and probability corresponding to each logging segment data; in particular, the amount of the solvent to be used,
Inputting logging segment data corresponding to each processed logging segment into a pre-trained deep neural network classification model, and realizing dimensionality reduction processing through multilayer network layer weight and bias and loading of an activation function to obtain processed logging data; the activation function is specifically shown in formula (2):
Wherein z isi+1is the output of the i-layer network layer; z is a radical ofiInputting logging segment data corresponding to each processed logging segment into an i-layer network layer; f. ofiAn activation function for an i-layer network layer; wiThe weight matrix is the weight matrix of the i-layer network layer; biThe offset vector of the i-layer network layer;
According to the obtained processed logging data, classifying a full connection layer of the last layer of the model by using a deep neural network and an activation function soft max, and according to a formula (3), obtaining the corresponding type and probability of each logging fragment data;
Wherein z ism,kthe original output of the full connection layer of the last layer; m is the last layer, k is the kth neuron of the last layer, zm,nFor the nth neuron of the last layer, sigmanexp(zm,n) Exponential form of all neurons in the last layer, soft max (z)m,k) The original output z of the fully-connected layer being the last layerm,kAnd (4) outputting after passing through the soft max function, namely, the probability of each logging segment data corresponding to the type.
If the target is marked as two types, two probabilities with different sizes can be output through a soft max function, the two probability values are compared, the larger probability value is selected as the final output and is marked as the probability corresponding to each logging segment data, the type corresponding to the larger probability is the type of the target and is marked as the type corresponding to each logging segment data.
The pre-trained deep neural network model is derived from the weight (W) of a deep neural network classification model obtained after supervised learning of a logging curve obtained based on numerical modelingi1,2, ….) and bias profile (b)i1,2, ….), including in particular:
According to the formula (4), constructing a framework of the deep neural network classification model:
y＝x1×x2+x1 (4)
Wherein, x1 and x2 respectively represent two attributes of the logging segment, namely x1 is a logging value in the logging segment after log-log and normalization processing, x2 is a multiple difference between the logging value on the logging segment and a mean square value of all logging values on the logging segment, namely a multiple feature of the logging segment, and y represents a judgment result of the type and probability of each logging segment realized according to the logging value of the logging segment and the multiple feature of the logging segment. The formula (4) embodies the framework idea of constructing the deep neural network classification model, and the response characteristics of the induction logging curve can be well extracted according to the idea.
Wherein MSE represents the mean square value of all logging values on a logging segment; m represents the number of logging values on the logging segment; ri,timeand (4) representing the multiple characteristic of the ith log value on the log fragment.
The architecture of the deep neural network classification model constructed according to equation 4, as shown in fig. 3, specifically includes: a first input layer, a second input layer, a third input layer, nine local connection layers, multiplied fusion layers, summed fusion layers, a transition layer, and a full connection layer;
the logging value of the logging segment and the multiple characteristics of the logging segment are used as input data and are respectively input into a first input layer, a second input layer and a third input layer, dimension reduction processing of the logging segment is realized through respective local connecting layers, combination of different attributes of the logging segment is realized through a multiplied fusion layer and an added fusion layer, and finally the type and the probability of the logging segment are obtained through a transition layer and a full connecting layer and through soft max.
the input of the deep neural network classification model is processed logging segment data, and the labeled value of a geological model adopted by numerical modeling is used as an output set;
specifically, the induction logging instrument is introduced to log response data of a Chirp geological model as a training set, and the Chirp model does not consider mud invasion situations. And setting the low resistance layer Rlow according to the measurement range (0.1-2000 ohmm) of induction logging as follows: 0.1ohmm, 0.2ohmm, 0.5ohmm, 1ohmm, 2ohmm, 5ohmm, 10ohmm, 20ohmm, 50ohmm, 100ohmm, 200ohmm, 500ohmm, 1000 ohmm. Resistivity ratio Rhigh/Rlow of the high resistance layer and the low resistance layer: 2,4,6,8, 10, 20, 40, 60, 80, 100, 200. And (4) totaling 111 sets of logging data of the Chirp model.
Log response values of a Chirp model are subjected to logarithm and normalization processing, multiple features are extracted to serve as an input set, a high-resistance layer of the Chirp model is marked as 1, a low-resistance layer of the Chirp model is marked as 0, and the Chirp model serves as an output set.
the training process adopts a cross entropy loss function, and weight (W) of each layer of the deep neural network model is obtained through repeated iterative computationi1,2, ….) and an offset (b)i,i＝1,2,….)。
The input set of the deep neural network classification model consisting of the processed logging segment data and the multiple characteristics of the logging values is multidimensional data, and reflects various attributes of a logging curve (the logging values reflect geological resistivity attributes corresponding to the logging depths, and the multiple characteristics of the logging values reflect stratum change characteristics corresponding to the logging depths); the local connecting layer gradually reduces the contribution of the edge data of the logging segment to the type and probability along with the increasing of the layer number, and the information dimension reduction processing of the logging segment is realized; the use of local tie layers also ensures the spatial sequence characteristic of the log, i.e., the spatial order of log values with respect to geological depth. In the training process, a training set of a plurality of contrast Chirp models is considered, the processing capacity of the layer interface identification method on the logging curve of the complex stratum is improved, and even the stratum with invasion characteristics can be processed.
And outputting the deep neural network classification model as the corresponding category and probability of the processed logging segment data.
Step 4) judging whether a layer interface exists between every two adjacent logging fragment data by comparing the corresponding type and probability of every two adjacent logging fragment data;
If the type and the probability corresponding to one logging segment data are different from those corresponding to the adjacent logging segment data, a layer interface exists between the two adjacent logging data, and the two adjacent logging data are identified and marked.
The method comprises the following steps that all processed logging segments are input into a pre-trained deep neural network classification model to obtain the type and probability of each logging segment, the type and probability represent the type and probability of a geological point corresponding to a logging depth of the logging segment, and layer interface marking of the geological model is realized by evaluating the type and probability of the geological point of two adjacent logging depths, specifically:
If the type and probability corresponding to the geological point with a certain logging depth are different from those corresponding to the geological point with the adjacent logging depth, judging that a layer interface exists between the adjacent geological points, and identifying and marking the layer interface so as to finish the layer interface identification of the logging curve.
The layer interface recognition effect of the Chirp model (Rhigh/Rlow ═ 100) is shown in fig. 4, and the processing effect of the deep neural network model can be explained. The horizontal coordinate is logging depth, the left coordinate is a target mark (0 or 1), the right coordinate is a resistivity value, o is a layer interface mark, the graph shows that the mark o is 1 on each layer interface of the Chirp model, the marks are zero at other positions, and the curve is the resistivity value of the Chirp model.
the Oklahoma geological model is a classical model for testing an electrical logging data processing method, and as shown in a figure 5a, the resistivity change range is 0.4-150.0 ohmm, the stratum thickness is 0.6-5.2 m, and the complex condition, the instrument response characteristic and the performance of a real stratum can be fully reflected. Wherein the solid line is the resistivity distribution of the Oklahoma geological model, and the dotted line is the induction log.
The layer interface identification method has good performance in thin layers around 20-32m, 56-58m, step layers around 51-54m and 37-42m, as shown in figure 5 b. The abscissa is the logging depth, the left coordinate is a mark (0 or 1), the right coordinate is a resistivity value, and o is a layer interface mark, and it is seen from the figure that the mark o is 1 on each layer interface of the Oklahoma model, and is zero at other positions, and the curve is the resistivity value of the Oklahoma model.
Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention and are not limited. Although the present invention has been described in detail with reference to the embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (5)
1. A layer interface identification method of an induction logging curve based on a deep neural network is characterized by comprising the following steps:
Windowing and cutting the logging curve acquired in real time to obtain a plurality of logging segments;
Carrying out logarithm and normalization processing on all logging values on each logging segment to obtain logging segment data corresponding to each processed logging segment;
inputting the acquired logging segment data corresponding to each processed logging segment into a pre-trained deep neural network classification model, and acquiring the type and probability corresponding to each logging segment data;
And judging whether a layer interface exists between every two adjacent logging fragment data or not by comparing the corresponding types and the corresponding probabilities of every two adjacent logging fragment data.
2. the method according to claim 1, wherein all the log values on each log segment are subjected to logarithmic and normalization processing to obtain log segment data corresponding to each processed log segment; the method specifically comprises the following steps:
According to the formula (1), all the logging values on each logging segment are subjected to logarithmic treatment and normalization treatment, and the logarithmic treatment and normalization treatment on all the logging values on each logging segment are completed:
wherein the content of the first and second substances,For the ith log value R on the log sliceiLog values after logarithmic and normalization processing;
obtaining all processed logging values on each logging segment according to the formula (1), and further obtaining each processed logging segment; thereby obtaining logging segment data corresponding to each processed logging segment; and the logging segment data corresponding to each processed logging segment is all logging values on each processed logging segment.
3. the method according to claim 1, wherein the obtained logging segment data corresponding to each processed logging segment is input into a pre-trained deep neural network classification model, and the category and probability corresponding to each logging segment data are obtained; the method specifically comprises the following steps:
inputting logging segment data corresponding to each processed logging segment into a pre-trained deep neural network classification model, and realizing dimensionality reduction processing through multilayer network layer weight and bias and loading of an activation function to obtain processed logging data:
zi+1＝fi(Wi T×zi+bi) (2)
Wherein z isi+1is the output of the i-layer network layer; z is a radical ofiis the input of the i-layer network layer; f. ofiAn activation function for an i-layer network layer; withe weight matrix is the weight matrix of the i-layer network layer; biThe offset vector of the i-layer network layer;
According to the obtained processed logging data, classifying a full connection layer of the last layer of the model by using a deep neural network and an activation function soft max, and according to a formula (3), obtaining the corresponding type and probability of each logging fragment data;
Wherein z ism,kThe original output of the full connection layer of the last layer; m is the last layer, k is the kth neuron of the last layer, zm,nFor the nth neuron of the last layer, sigmanexp(zm,n) Exponential form of all neurons in the last layer, soft max (z)m,k) The original output z of the fully-connected layer being the last layerm,koutputting after passing through the soft max function, namely, the probability of each corresponding type of the logging segment data;
if the target is marked as two types, two probabilities with different sizes can be output through a soft max function, the two probability values are compared, the larger probability value is selected as the final output and is marked as the probability corresponding to each logging segment data; the category corresponding to the higher probability is the category of the target and is recorded as the category corresponding to each logging segment data.
4. the method according to claim 3, wherein the deep neural network classification model specifically comprises: a first input layer, a second input layer, a third input layer, a plurality of local connection layers, a multiplied fusion layer, an added fusion layer, a transition layer, and a full connection layer;
The logging value of the logging segment and the multiple characteristics of the logging segment are used as input data and are respectively input into a first input layer, a second input layer and a third input layer, dimension reduction processing of the logging segment is realized through respective local connecting layers, combination of different attributes of the logging segment is realized through a multiplied fusion layer and an added fusion layer, and finally the type and the probability of the logging segment are obtained through a transition layer and a full connecting layer and through soft max.
5. The method according to claim 4, wherein the judging whether there is a layer interface between two adjacent log fragment data by comparing the type and probability corresponding to each two adjacent log fragment data is specifically:
If the type and the probability corresponding to one logging segment data are different from those corresponding to the adjacent logging segment data, a layer interface exists between the two adjacent logging data, and the two adjacent logging data are identified and marked.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910886031.4A CN110552693A (en) | 2019-09-19 | 2019-09-19 | layer interface identification method of induction logging curve based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910886031.4A CN110552693A (en) | 2019-09-19 | 2019-09-19 | layer interface identification method of induction logging curve based on deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110552693A true CN110552693A (en) | 2019-12-10 |
Family
ID=68740775
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910886031.4A Pending CN110552693A (en) | 2019-09-19 | 2019-09-19 | layer interface identification method of induction logging curve based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110552693A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111582560A (en) * | 2020-04-22 | 2020-08-25 | 空间信息产业发展股份有限公司 | Rice yield prediction method based on recurrent neural network |
| CN112348922A (en) * | 2020-11-06 | 2021-02-09 | 长江大学 | Method, system and device for automatically drawing logging curve and storage medium |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101630013A (en) * | 2008-07-16 | 2010-01-20 | 中国石油天然气集团公司 | Method for inverting Poisson ratio parameters of pre-stack seismic data |
| CN102053270A (en) * | 2009-10-30 | 2011-05-11 | 中国石油化工股份有限公司 | Sedimentary formation unit-based seismic facies analysis method |
| CN102418518A (en) * | 2011-04-12 | 2012-04-18 | 北京师范大学 | Method for identifying water logging grades of oil reservoir by using neural network analogue cross plot |
| CN202946122U (en) * | 2012-12-20 | 2013-05-22 | 北京盛锐马科技有限公司 | Shock absorption system of petroleum drilling and logging device |
| CN104992178A (en) * | 2015-06-16 | 2015-10-21 | 中国石油化工股份有限公司 | Tight sandstone fluid type identification method based on support vector machine simulation cross plot |
| CN106504215A (en) * | 2016-11-10 | 2017-03-15 | 中国科学院长春光学精密机械与物理研究所 | A kind of method and device of image enhaucament |
| CN106991509A (en) * | 2017-05-27 | 2017-07-28 | 重庆科技学院 | Log Forecasting Methodology based on radial basis function neural network model |
| CN109709603A (en) * | 2018-11-23 | 2019-05-03 | 中国石油天然气股份有限公司 | Seismic horizon identification and method for tracing, system |
| CN109736790A (en) * | 2018-12-28 | 2019-05-10 | 新疆工程学院 | Recognition of Weil Logging Lithology method based on deepness belief network |
| CN109857979A (en) * | 2018-12-29 | 2019-06-07 | 中国船舶重工集团公司第七一八研究所 | A kind of log approximating method based on wavelet analysis and BP neural network |
-
2019
- 2019-09-19 CN CN201910886031.4A patent/CN110552693A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101630013A (en) * | 2008-07-16 | 2010-01-20 | 中国石油天然气集团公司 | Method for inverting Poisson ratio parameters of pre-stack seismic data |
| CN102053270A (en) * | 2009-10-30 | 2011-05-11 | 中国石油化工股份有限公司 | Sedimentary formation unit-based seismic facies analysis method |
| CN102418518A (en) * | 2011-04-12 | 2012-04-18 | 北京师范大学 | Method for identifying water logging grades of oil reservoir by using neural network analogue cross plot |
| CN202946122U (en) * | 2012-12-20 | 2013-05-22 | 北京盛锐马科技有限公司 | Shock absorption system of petroleum drilling and logging device |
| CN104992178A (en) * | 2015-06-16 | 2015-10-21 | 中国石油化工股份有限公司 | Tight sandstone fluid type identification method based on support vector machine simulation cross plot |
| CN106504215A (en) * | 2016-11-10 | 2017-03-15 | 中国科学院长春光学精密机械与物理研究所 | A kind of method and device of image enhaucament |
| CN106991509A (en) * | 2017-05-27 | 2017-07-28 | 重庆科技学院 | Log Forecasting Methodology based on radial basis function neural network model |
| CN109709603A (en) * | 2018-11-23 | 2019-05-03 | 中国石油天然气股份有限公司 | Seismic horizon identification and method for tracing, system |
| CN109736790A (en) * | 2018-12-28 | 2019-05-10 | 新疆工程学院 | Recognition of Weil Logging Lithology method based on deepness belief network |
| CN109857979A (en) * | 2018-12-29 | 2019-06-07 | 中国船舶重工集团公司第七一八研究所 | A kind of log approximating method based on wavelet analysis and BP neural network |
Non-Patent Citations (2)
| Title |
|---|
| 许宏龙: "基于BP神经网络技术的储层非均质性研究", 《中国优秀硕士学位论文全文数据库 基础科学辑》 * |
| 韩东伟等: "基于深度学习网络的烟叶质量识别", 《安徽农业科学》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111582560A (en) * | 2020-04-22 | 2020-08-25 | 空间信息产业发展股份有限公司 | Rice yield prediction method based on recurrent neural network |
| CN111582560B (en) * | 2020-04-22 | 2023-05-23 | 空间信息产业发展股份有限公司 | Rice yield prediction method based on cyclic neural network |
| CN112348922A (en) * | 2020-11-06 | 2021-02-09 | 长江大学 | Method, system and device for automatically drawing logging curve and storage medium |
| CN112348922B (en) * | 2020-11-06 | 2023-04-07 | 长江大学 | Method, system and device for automatically drawing logging curve and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3437067B1 (en) | Automated core description | |
| Wang et al. | Marcellus shale lithofacies prediction by multiclass neural network classification in the Appalachian Basin | |
| WO2017084454A1 (en) | Stratum component optimization determination method and device | |
| US20220237891A1 (en) | Method and system for image-based reservoir property estimation using machine learning | |
| CN114723095A (en) | Missing well logging curve prediction method and device | |
| CN110552693A (en) | layer interface identification method of induction logging curve based on deep neural network | |
| WO2022159698A1 (en) | Method and system for image-based reservoir property estimation using machine learning | |
| CN116168224A (en) | Machine learning lithology automatic identification method based on imaging gravel content | |
| CN111626377B (en) | Lithology recognition method, device, equipment and storage medium | |
| WO2021163582A1 (en) | Systems and methods for generating continuous grain size logs from petrographic thin section images | |
| CN117093922A (en) | Improved SVM-based complex fluid identification method for unbalanced sample oil reservoir | |
| Sarkheil et al. | The fracture network modeling in naturally fractured reservoirs using artificial neural network based on image loges and core measurements | |
| CN117131971A (en) | Reservoir dominant seepage channel prediction method based on Xgboost algorithm | |
| CN115809411A (en) | Improved decision tree lithology identification method based on logging data | |
| CN114991757A (en) | Formation pressure measurement and recording integrated monitoring method and system while drilling | |
| CN112147679B (en) | Lithology prediction method and device based on elastic parameters under fuzzy logic framework | |
| Dai et al. | Stratigraphic automatic correlation using SegNet semantic segmentation model | |
| CN117174203B (en) | Logging curve response analysis method for sandstone uranium ores | |
| CN113050191A (en) | Shale oil TOC prediction method and device based on double parameters | |
| CN110674868A (en) | Stratum lithology identification system and method based on high-dimensional drilling parameter information | |
| CN112906242B (en) | Geophysical modeling method based on combination of naive Bayes method and proximity classification method | |
| CN115576028B (en) | Geological feature layer prediction method and system based on support vector machine | |
| de Oliveira et al. | Ensemble of heterogeneous classifiers applied to lithofacies classification using logs from different wells | |
| Soleimani et al. | Classification of rock facies using deep convolutional neural network | |
| CN116187206B (en) | COD spectrum data migration method based on generation countermeasure network |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191210 |
|
| RJ01 | Rejection of invention patent application after publication |