CN114114414A - Artificial intelligence prediction method for 'dessert' information of shale reservoir - Google Patents
Artificial intelligence prediction method for 'dessert' information of shale reservoir Download PDFInfo
- Publication number
- CN114114414A CN114114414A CN202111369371.3A CN202111369371A CN114114414A CN 114114414 A CN114114414 A CN 114114414A CN 202111369371 A CN202111369371 A CN 202111369371A CN 114114414 A CN114114414 A CN 114114414A
- Authority
- CN
- China
- Prior art keywords
- data
- prediction
- training
- parameters
- seismic data
- 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
- 238000000034 method Methods 0.000 title claims abstract description 54
- 235000021185 dessert Nutrition 0.000 title claims abstract description 20
- 238000013473 artificial intelligence Methods 0.000 title claims abstract description 12
- 238000012549 training Methods 0.000 claims abstract description 60
- 235000009508 confectionery Nutrition 0.000 claims abstract description 27
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 20
- 238000004458 analytical method Methods 0.000 claims abstract description 8
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000004140 cleaning Methods 0.000 claims abstract description 7
- 238000011156 evaluation Methods 0.000 claims abstract description 6
- 238000010606 normalization Methods 0.000 claims description 22
- 238000012795 verification Methods 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 7
- 239000011435 rock Substances 0.000 claims description 7
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 4
- 230000035945 sensitivity Effects 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 150000001875 compounds Chemical class 0.000 claims description 3
- 238000009826 distribution Methods 0.000 abstract description 9
- 230000008901 benefit Effects 0.000 abstract description 6
- 238000012512 characterization method Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 16
- 239000007789 gas Substances 0.000 description 13
- 230000008569 process Effects 0.000 description 7
- 230000000694 effects Effects 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 4
- 239000004215 Carbon black (E152) Substances 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 229930195733 hydrocarbon Natural products 0.000 description 3
- 150000002430 hydrocarbons Chemical class 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 230000003247 decreasing effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 239000003345 natural gas Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 241001532036 Aspidistra Species 0.000 description 1
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 102100027611 Rho-related GTP-binding protein RhoB Human genes 0.000 description 1
- 101150054980 Rhob gene Proteins 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000009933 burial Methods 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000994 depressogenic effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000008717 functional decline Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000005416 organic matter Substances 0.000 description 1
- 230000010363 phase shift Effects 0.000 description 1
- 239000011148 porous material Substances 0.000 description 1
- 238000001179 sorption measurement Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
- G01V1/30—Analysis
- G01V1/306—Analysis for determining physical properties of the subsurface, e.g. impedance, porosity or attenuation profiles
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
- G01V1/282—Application of seismic models, synthetic seismograms
Landscapes
- Engineering & Computer Science (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Acoustics & Sound (AREA)
- Environmental & Geological Engineering (AREA)
- Geology (AREA)
- General Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Geophysics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an artificial intelligence prediction method for 'dessert' information of a shale reservoir, which comprises the following steps: step 1, preprocessing data: cleaning, normalizing and equalizing the logging data and the seismic data; step 2, constructing a convolutional neural network: determining a network structure according to the input characteristics and the form of an output result; step 3, training a dessert parameter prediction model based on the logging data; and 4, three-dimensional prediction is carried out on 'sweet spot' parameters based on the seismic data to generate a data volume, and the shale 'sweet spot' distribution condition is evaluated by utilizing the analysis result of professional seismic data visualization software. The prediction of the 'sweet spot' parameters TOC, PHI and GAS of the reservoir is realized by utilizing the one-dimensional convolutional neural network, the advantages of strong CNN nonlinear characterization capability and difficulty in falling into a local optimal solution are fully utilized, the 'sweet spot' parameter prediction precision is greatly improved, and a basis can be provided for the prediction evaluation of the 'sweet spot' of the shale GAS.
Description
Technical Field
The invention belongs to the field of geoscience and the field of machine learning, and particularly relates to an artificial intelligence prediction method for 'dessert' information of a shale reservoir.
Background
Shale gas is a natural gas resource stored in shale reservoirs, existing in natural fractures and pores in free, adsorbed, etc. form. The reserves of the shale gas in China are in the front of all countries in the world, the demand of the shale gas for energy is large in China, and the shale gas has good development prospect. However, the research on shale gas exploration and development in China starts late, and the shale gas distribution in China has the characteristics of deep reservoir burial, low abundance, complex and various geological structures, and is difficult to explore and develop and high in cost, so that the large-scale development of the shale gas is far from the priority.
The shale gas dessert is an area or a layer which is enriched and easy to develop in shale gas phase, has high economic benefit, and is particularly important for accurately predicting and identifying the shale dessert. The key parameters of the shale GAS geological dessert comprise total organic matter content (TOC), GAS content (GAS), Porosity (PHI) and the like, and the accurate parameter values are obtained, so that the method has important significance for identification and prediction of the shale GAS dessert. Methods for directly obtaining such parameter values are core sample analysis, which has the limitations of high sampling cost, long cycle length, often little and discontinuous data, and sample instability during each sampling and analysis may introduce errors.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides an artificial intelligent prediction method for 'sweet spot' information of a shale reservoir, which uses CNN to realize the prediction of TOC, PHI and GAS parameters and generates a seismic data volume to evaluate the shale GAS sweet spot distribution condition in a research area.
The purpose of the invention is realized by the following technical scheme: an artificial intelligence prediction method for 'dessert' information of a shale reservoir comprises the following steps:
step 1, preprocessing data: cleaning, normalizing and equalizing the logging data and the seismic data;
step 3, training a dessert parameter prediction model based on the logging data;
and 4, three-dimensional prediction is carried out on the 'sweet spot' parameters based on the seismic data to generate a data volume, and the data volume is input into the prediction model trained in the step 3 for prediction.
Further, the specific implementation method of step 1 is as follows:
step 11, carrying out data cleaning on the logging data and the seismic data, removing null values and abnormal values, and selecting the logging data in the same stratum range as the seismic data as a sample; the logging data is used for training and verifying a prediction model, and the seismic data is data to be predicted;
wherein V, K, E represents Poisson's ratio, bulk modulus, Young's modulus, and ρ, Vp, Vs represent density, longitudinal wave velocity, and transverse wave velocity, respectively;
step 13, data normalization: when the sample contains multiple logging data, a method of firstly standardizing single logging data and then carrying out the most-value normalization on all logging data is adopted:
in the formula, xstdIs a normalized value, xscaledIs the value after the most value normalization, x,
σ、xmaxAnd xminRespectively obtaining an original numerical value, a sample mean value, a sample standard deviation, a sample minimum value and a sample maximum value;
and 14, carrying out equalization treatment on the sample data to ensure that the reservoir data accounts for not less than 20%.
Further, the step 2 is specifically implemented as follows:
step 21, constructing a network forward propagation basic framework: adopting a one-dimensional convolution neural network with 4 input features, wherein the 4 input features firstly pass through two one-dimensional convolution layers and respectively comprise 64 and 32 one-dimensional convolution kernels with the length of 2 to obtain 32 high-level feature maps with the length of 2; then, passing through a maximum pooling layer to reduce the characteristic length by half, wherein the characteristic length is 1; then passing through a flattening layer and a full connecting layer containing 15 units to finally reach an output layer;
the one-dimensional convolutional layer and the full-connection layer both adopt a ReLU activation function, and the output layer uses a Sigmoid activation function;
step 22, determining a network loss function and a back propagation algorithm, wherein the loss function uses an average absolute value error:
where MAE represents the mean absolute error, n is the number of samples, yiIndicates the actual value of the ith sample,
representing the ith sample prediction value; selecting an Adam optimizer for the weight updating algorithm;
step 23, evaluating the network output precision: by means of R2Index evaluation regression task accuracy:
in the formula (I), the compound is shown in the specification,
represents the sample mean; if R is2And when the index meets the preset requirement, the prediction accuracy of the model meets the requirement, and the training is ended, otherwise, the parameters are adjusted to perform the training again.
Further, the specific implementation method of step 3 is as follows:
step 31, performing 'sweet spot' parameter sensitivity characteristic analysis according to the acquired logging data, drawing an intersection graph of TOC, PHI and GAS parameters about density, wave velocity, Poisson ratio, volume modulus and Young modulus, and determining a characteristic set; forming a sample by the feature set and the target prediction parameters;
step 32, dividing the samples into a training set and a verification set, wherein in each iteration, 80% of samples in the training set are used for training, and the rest 20% of samples in the training set are used for verification; the training of the model is completed on a Tensorflow platform, an early stopping point is set in the training, and the training is stopped when the loss function tends to be stable and the verification precision meets the requirement;
step 33, the training set and the verification set are divided again, and the training model is repeated to verify the stability; and storing the prediction model after the accuracy and stability requirements are met.
Further, the specific implementation method of step 4 is as follows:
step 41, preprocessing of seismic data: calculating corresponding seismic data rock mechanical parameters for each seismic channel according to the characteristic combination determined in the step 31; standardizing the seismic data, and then carrying out the most value normalization to keep the seismic data consistent with the logging data;
42, inputting the preprocessed seismic data feature set into a prediction model for prediction, performing inverse normalization on an output result, and performing inverse normalization to obtain a final prediction result;
and 43, writing the prediction result into a seismic channel corresponding to the sgy-format seismic data to generate a three-dimensional data body of the 'dessert' parameter, and analyzing the result by using professional seismic data visualization software.
The invention has the beneficial effects that: the prediction of 'dessert' parameters TOC, PHI and GAS of the reservoir is realized by utilizing the one-dimensional convolutional neural network (1D-CNN), and the advantages of strong nonlinear characterization capability of the CNN and difficulty in falling into a local optimal solution are fully utilized. The model is compatible with the prediction of depth domain logging data and time domain seismic data, and the characteristic information loss caused by depth-time resampling of the logging data is avoided. The model can realize the prediction of various parameters by using different characteristic combinations, and has good stability and strong generalization capability. Meanwhile, different preprocessing methods are provided for the characteristics of the logging data, single-port logging data are standardized, and then the most-valued normalization is carried out on the multiple-port logging data, so that the prediction precision of the 'sweet spot' parameter is greatly improved. For different input characteristic combinations, the prediction of the three parameters has a better effect, so that the prediction method has certain universality and robustness. The method has the advantages that the method also has a good effect in the three-dimensional prediction of the 'sweet spot' parameters based on the seismic data, and areas (horizons) with high TOC, PHI and GAS values are consistent with the known values, so that the method can provide a basis for the identification and prediction of the 'sweet spot' of the unknown shale GAS geology.
Drawings
FIG. 1 is a flow chart of shale reservoir "sweet spot" parameter prediction based on a convolutional neural network according to the present invention;
FIG. 2 is a "sweet-spot" parameter prediction model with 4 input features;
FIG. 3 is a graph of the interaction of GAS parameters with density, compressional velocity, and shear velocity;
FIG. 4 is a plot of the GAS parameters versus Poisson's ratio, bulk modulus, Young's modulus;
FIG. 5 is a loss function descent curve and an R-Square change curve for the PHI parameter prediction model training process;
FIG. 6 is a comparison of the PHI parameter prediction result and the true value;
FIG. 7 is a loss function descent curve and an R-Square variation curve for the TOC parameter prediction model training process;
FIG. 8 is a comparison of TOC parameter prediction results and true values;
FIG. 9 is a loss function descent curve and an R-Square change curve for the GAS parameter prediction model training process;
FIG. 10 is a comparison of the GAS parameter prediction results with the true values;
FIG. 11 is a cross-sectional view of a three-dimensional prediction of TOC parameters;
FIG. 12 is a three-dimensional predicted profile of the PHI parameters;
FIG. 13 is a three-dimensional predicted profile of the GAS parameter;
FIG. 14 is a cross-sectional view of a TOC parameter three-dimensional predicted target reservoir;
FIG. 15 is a section view of a PHI parameter three-dimensional prediction target reservoir;
FIG. 16 is a cross-sectional view of a GAS parameter three-dimensional predicted target reservoir.
Detailed Description
The logging information can reflect longitudinal lithology and horizon change information of the well position, has the advantages of continuity and high longitudinal resolution compared with core data, and shale gas sweet-spot key parameter values such as TOC and the like are obtained through analysis of hydrocarbon source rock sensitive parameters such as acoustic time difference (AC), density (RHOB), natural Gamma (GR) and the like. The existing logging parameter prediction methods include a delta LogR method, a natural gamma energy spectrum logging method and the like, and supervised learning methods such as multiple linear regression, support vector regression, decision tree regression, neural network and the like.
Compared with the traditional prediction method and other machine learning methods, the neural network has outstanding nonlinear advantages, and can realize complex nonlinear mapping between input characteristic parameters and output parameters. In well logging and seismic attribute analysis, due to the heterogeneity and complexity of the formation, the measured parameter values are affected by various factors, and the feature mapping relationship between the input parameters and the target prediction parameters is very complex. Therefore, neural networks have great potential in regression prediction of curves. The Convolutional Neural Network (CNN) further introduces nonlinearity through operations of convolution and pooling, and is less prone to fall into a locally optimal solution than a BP network; meanwhile, the method of convolutional layer sparse connection and weight sharing reduces the size of the network and reduces the training pressure of the network. Therefore, prediction of TOC, PHI and GAS parameters can be achieved using CNN, generating seismic data volumes to evaluate shale GAS sweet spot distribution within the area of interest.
The technical scheme of the invention is further explained by combining the attached drawings.
As shown in fig. 1, the artificial intelligence prediction method for the 'sweet spot' information of the shale reservoir provided by the invention comprises the following steps:
step 1, preprocessing data: cleaning, normalizing and equalizing the logging data and the seismic data; the specific implementation method comprises the following steps:
step 11, carrying out data cleaning on the logging data and the seismic data, removing null values and abnormal values, and selecting the logging data in the same stratum range as the seismic data as a sample; the logging data is used for training and verifying a prediction model, and the seismic data is data to be predicted;
wherein V, K, E represents Poisson's ratio, bulk modulus, Young's modulus, and ρ, Vp, Vs represent density, longitudinal wave velocity, and transverse wave velocity, respectively;
step 13, data normalization: when the sample contains multiple logging data, a method of firstly standardizing single logging data and then carrying out the most-value normalization on all logging data is adopted:
in the formula, xstdIs a normalized value, xscaledIs the value after the most value normalization, x,
σ、xmaxAnd xminRespectively obtaining an original numerical value, a sample mean value, a sample standard deviation, a sample minimum value and a sample maximum value; the data normalized by the method can be used as the input of the convolutional neural network to ensure the normal operation of network training. Meanwhile, storing the standardization and normalization parameters for the inverse normalization of the output result;
and step 14, carrying out equalization treatment on sample data, wherein the shale GAS reservoir is an unconventional natural GAS reservoir in which shale GAS is mainly in an adsorption state and a free state and a small amount of dissolved state exists in shale, and the shale GAS reservoir has the characteristics of high PHI (phase shift indicator), TOC (Total organic carbon) and GAS (GAS area) values in logging parameters. The effective logging data depth range can reach hundreds of meters, and the shale gas reservoir thickness is generally small. If the non-reservoir proportion in the sample data is too large, the prediction effect of the model obtained by training on reservoir parameters can be influenced. Therefore, the logging data of the non-reservoir stratum is properly reduced to reach the proper proportion between the reservoir stratum data and the non-reservoir stratum data, and the reservoir stratum data accounts for not less than 20%.
step 21, constructing a network forward propagation basic framework: as shown in fig. 2, a one-dimensional convolutional neural network (1D-CNN model) with 4 input features is adopted, the 4 input features first pass through two one-dimensional convolutional layers, and respectively include 64 and 32 one-dimensional convolutional kernels with the length of 2, so as to obtain 32 high-level feature maps with the length of 2; then, passing through a maximum pooling layer to reduce the characteristic length by half, wherein the characteristic length is 1; then passing through a flattening layer and a full connection layer containing 15 units to finally reach an output layer corresponding to TOC, PHI, GAS or other target prediction parameters;
the one-dimensional convolutional layer and the full-connection layer adopt a ReLU activation function: (x) max {0, x }, the output layer uses the Sigmoid activation function:
step 22, determining a network loss function and a back propagation algorithm, wherein the loss function uses an average absolute value error:
where MAE represents the mean absolute error, n is the number of samples, yiIndicates the actual value of the ith sample,
representing the ith sample prediction value; the gradient of the MAE does not decrease with decreasing error, so the learning rate is reduced appropriately with decreasing error in coordination with the dynamic learning rate. Selecting an Adam optimizer for the weight updating algorithm;
step 23, evaluating the network output precision: by means of R2Index evaluation regression task accuracy:
in the formula (I), the compound is shown in the specification,
represents the sample mean; if R is2And when the index meets the preset requirement, the prediction accuracy of the model meets the requirement, and the training is ended, otherwise, the parameters are adjusted to perform the training again. R2A maximum value of 1, with closer to 1 indicating higher accuracy of the regression prediction. In general, if R2If the index is greater than 0.8, the prediction is better, and the result is reliable; if R is2If the index is less than 0.5, the accuracy of the model is insufficient, and the parameters need to be adjusted or the model and the equation need to be reconsideredThe method is carried out. The specific preset criteria may be determined in combination with the sample data used.
Step 3, training a dessert parameter prediction model based on the logging data; the sensitive characteristics of the TOC, PHI, and GAS parameters are first analyzed based on the well log data to determine the network inputs. And in the training process, a cross validation method is used for evaluating the accuracy of the model, and the stability of the model is verified through multiple times of training.
The specific implementation method comprises the following steps:
step 31, performing 'sweet spot' parameter sensitivity characteristic analysis according to the acquired logging data, drawing an intersection graph of TOC, PHI and GAS parameters about density, wave velocity, Poisson ratio, volume modulus and Young modulus, and determining a characteristic set; forming a sample by the feature set and the target prediction parameters;
step 32, dividing the samples into a training set and a verification set, wherein in each iteration, 80% of samples in the training set are used for training, and the rest 20% of samples in the training set are used for verification; the validation set was used for final evaluation and did not participate in model training. The training of the model is completed on a Tensorflow platform, and the training speed is improved and overfitting is inhibited by utilizing technologies such as Batch Normalization and Dropout. Setting Early Stopping (Early Stopping) points in training, and Stopping training when the loss function tends to be stable and the verification precision meets the requirements;
step 33, the training set and the verification set are divided again, and the training model is repeated to verify the stability; and storing the prediction model after the accuracy and stability requirements are met.
step 41, preprocessing of seismic data: calculating corresponding seismic data rock mechanical parameters for each seismic channel according to the characteristic combination determined in the step 31; standardizing the seismic data, and then carrying out the most value normalization to keep the seismic data consistent with the logging data;
42, inputting the preprocessed seismic data feature set into a prediction model for prediction, performing inverse normalization on an output result, and performing inverse normalization to obtain a final prediction result;
and 43, writing the prediction result into a seismic channel corresponding to the sgy-format seismic data to generate a three-dimensional data body of the 'dessert' parameter, and analyzing the result by using professional seismic data visualization software.
Example 1 "sweet spot" parameter prediction model training and validation based on well log data
In order to verify the accuracy and stability of the shale reservoir 'sweet spot' information artificial intelligence prediction method model, firstly, the model is trained and verified by utilizing logging data. Logging data are from 4 logs of a Weirong shale gas field depressed in the west of the Sichuan basin, and the depth range is 3270-3850 meters.
Fig. 3(a) to 3(c) are graphs of the GAS parameters with density, longitudinal wave velocity, and transverse wave velocity, respectively. The GAS parameters and the density have stronger negative correlation, and the reservoir and non-reservoir are obviously distinguished; the GAS has no obvious linear relation with the compressional wave velocity, but the reservoir and the non-reservoir are distinguished; GAS and shear wave velocity have no obvious distribution rule in the cross plot, but may have some complex and implicit characteristic rules.
Fig. 4(a) to 4(c) are graphs of the GAS parameter as a function of poisson's ratio, bulk modulus, and young's modulus, respectively. It can be seen that GAS has a certain degree of negative correlation with poisson ratio; in a GAS and bulk modulus intersection graph, the separability of a reservoir and a non-reservoir with respect to the bulk modulus is better; GAS has no significant characteristic relationship with young's modulus, and young's modulus is therefore not used. In conclusion, four parameters of density, longitudinal wave velocity, poisson ratio and volume modulus are selected as input characteristics of the GAS prediction model. Sensitivity profiles for TOC, PHI parameters have similar results to GAS, so the same input profile combinations are used.
The 1D-CNN model shown in FIG. 2 is used for training, and a loss function decline curve and an R-Square change condition during training are recorded.
Fig. 5(a) is a loss drop curve of PHI parametric model training, with model convergence after 120 iterations. FIG. 5(b) shows the R-Square variation of the training process, which eventually approaches 1. FIG. 6 is a comparison of the prediction result of the PHI parameter prediction model and the true value, which shows that the logging curve is basically consistent and the R-Square index reaches 0.99.
FIG. 7(a) is a loss drop curve for TOC parametric model training, with model convergence after 125 iterations. FIG. 7(b) shows the R-Square variation of the training process, eventually approaching 1. FIG. 8 is a comparison of the prediction results of the TOC parameter prediction model with the true values, which shows that the well logging curves are basically consistent, and the R-Square index is also 0.99.
FIG. 9(a) is a loss drop curve for GAS parametric model training, with model convergence after 140 iterations. FIG. 9(b) shows the R-Square variation of the training process, eventually approaching 1. FIG. 10 is a comparison of the prediction results of the GAS parameter prediction model and the true values, and it can be seen that the well logging curve fitting degree is good, and the R-Square index can reach 0.97.
In conclusion, the TOC, PHI and GAS curves of the well can be accurately predicted according to the density, the longitudinal wave velocity, the Poisson ratio and the bulk modulus by using the method. The prediction capability of the model on PHI and TOC is stronger than that of GAS, and good fitting effect can be obtained more easily because the negative correlation relationship between PHI and TOC and density parameters is strong. GAS is affected by more factors, so the prediction accuracy is slightly lower than TOC and PHI. Comparing the results obtained by the method with other machine learning methods such as table 1, the method can obtain better fitting effect compared with BP neural network (BP), Support Vector Regression (SVR) and K nearest neighbor regression (KNN) methods, has the highest R-Square and the lowest MAE and MSE values in the prediction of TOC, PHI and GAS, and has high prediction accuracy. In conclusion, the 1D-CNN prediction method is high in precision and strong in model stability, can be used for predicting different parameters, and has reliability.
TABLE 1
Example 2 three-dimensional prediction of "sweet-spot" parameters based on seismic data
Three-dimensional prediction of the "sweet spot" parameter distribution was performed using seismic data based on the prediction model trained in example 1. The seismic data originated from a certain region of the Weirong shale gas field, containing 419971 seismic traces. And taking the density, longitudinal wave velocity, Poisson's ratio and volume modulus data of each seismic channel as input, and corresponding to the input of the prediction model. And respectively predicting TOC, PHI and GAS parameters to generate a three-dimensional data volume.
FIGS. 11-13 are cross-sectional views of the TOC, PHI, GAS parameters of the predicted results, respectively. The reservoir is known as the aspidistra ramulis ramstream group, is positioned around 3849 meters underground WY23, and corresponds to the seismic data time domain depth of 1.89 s. In the section diagrams of the three parameters, the target storage layer has obvious high value, the thickness is about 20m, and the transverse continuity is good. High TOC values are better hydrocarbon producing conditions, high PHI values favor hydrocarbon storage, provide conditions for high GAS values, and the results are consistent.
FIGS. 14-16 are layer diagrams of the prediction results TOC, PHI, and GAS parameters, respectively. The distribution of the parameters GAS shows certain similarity with TOC and PHI, and accords with the geoscience law. The GAS values of the target reservoirs around the WY23, WY29 and WY35 wells are higher, and the GAS production capacity is better. Therefore, the 1D-CNN prediction method can realize three-dimensional prediction of 'sweet spot' parameters based on seismic data on the basis of model training and prediction based on logging data, accurately depict the distribution situation of the parameters in spatial dimension, and provide a new method for prediction and evaluation of 'sweet spots' of shale reservoirs.
It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.
Claims (5)
1. An artificial intelligence prediction method for 'dessert' information of a shale reservoir is characterized by comprising the following steps:
step 1, preprocessing data: cleaning, normalizing and equalizing the logging data and the seismic data;
step 2, constructing a convolutional neural network: determining a network structure according to the input characteristics and the form of an output result;
step 3, training a dessert parameter prediction model based on the logging data;
and 4, three-dimensional prediction is carried out on the 'sweet spot' parameters based on the seismic data to generate a data volume, and the data volume is input into the prediction model trained in the step 3 for prediction.
2. The artificial intelligence prediction method for the 'sweet spot' information of the shale reservoir as claimed in claim 1, wherein the specific implementation method of step 1 is as follows:
step 11, carrying out data cleaning on the logging data and the seismic data, removing null values and abnormal values, and selecting the logging data in the same stratum range as the seismic data as a sample; the logging data is used for training and verifying a prediction model, and the seismic data is data to be predicted;
step 12, calculating rock mechanical parameters according to the existing logging parameters:
wherein V, K, E represents Poisson's ratio, bulk modulus, Young's modulus, and ρ, Vp, Vs represent density, longitudinal wave velocity, and transverse wave velocity, respectively;
step 13, data normalization: when the sample contains multiple logging data, a method of firstly standardizing single logging data and then carrying out the most-value normalization on all logging data is adopted:
in the formula, xstdIs a normalized value, xscaledIs the value after the most value normalization, x,
σ、xmaxAnd xminRespectively obtaining an original numerical value, a sample mean value, a sample standard deviation, a sample minimum value and a sample maximum value;
and 14, carrying out equalization treatment on the sample data to ensure that the reservoir data accounts for not less than 20%.
3. The artificial intelligence prediction method for the 'sweet spot' information of the shale reservoir as claimed in claim 1, wherein the step 2 is realized by the following steps:
step 21, constructing a network forward propagation basic framework: adopting a one-dimensional convolution neural network with 4 input features, wherein the 4 input features firstly pass through two one-dimensional convolution layers and respectively comprise 64 and 32 one-dimensional convolution kernels with the length of 2 to obtain 32 high-level feature maps with the length of 2; then, passing through a maximum pooling layer to reduce the characteristic length by half, wherein the characteristic length is 1; then passing through a flattening layer and a full connecting layer containing 15 units to finally reach an output layer;
the one-dimensional convolutional layer and the full-connection layer both adopt a ReLU activation function, and the output layer uses a Sigmoid activation function;
step 22, determining a network loss function and a back propagation algorithm, wherein the loss function uses an average absolute value error:
where MAE represents the mean absolute error, n is the number of samples, yiIndicates the actual value of the ith sample,
representing the ith sample prediction value; selecting an Adam optimizer for the weight updating algorithm;
step 23, evaluating the network output precision: by means of R2Index evaluation regression task accuracy:
in the formula (I), the compound is shown in the specification,
represents the sample mean; if R is2And when the index meets the preset requirement, the prediction accuracy of the model meets the requirement, and the training is ended, otherwise, the parameters are adjusted to perform the training again.
4. The artificial intelligence prediction method for the 'sweet spot' information of the shale reservoir as claimed in claim 1, wherein the step 3 is realized by the following steps:
step 31, performing 'sweet spot' parameter sensitivity characteristic analysis according to the acquired logging data, drawing an intersection graph of TOC, PHI and GAS parameters about density, wave velocity, Poisson ratio, volume modulus and Young modulus, and determining a characteristic set; forming a sample by the feature set and the target prediction parameters;
step 32, dividing the samples into a training set and a verification set, wherein in each iteration, 80% of samples in the training set are used for training, and the rest 20% of samples in the training set are used for verification; the training of the model is completed on a Tensorflow platform, an early stopping point is set in the training, and the training is stopped when the loss function tends to be stable and the verification precision meets the requirement;
step 33, the training set and the verification set are divided again, and the training model is repeated to verify the stability; and storing the prediction model after the accuracy and stability requirements are met.
5. The artificial intelligence prediction method for the dessert information of the shale reservoir as claimed in claim 4, wherein the step 4 is realized by the following steps:
step 41, preprocessing of seismic data: calculating corresponding seismic data rock mechanical parameters for each seismic channel according to the characteristic combination determined in the step 31; standardizing the seismic data, and then carrying out the most value normalization to keep the seismic data consistent with the logging data;
42, inputting the preprocessed seismic data feature set into a prediction model for prediction, performing inverse normalization on an output result, and performing inverse normalization to obtain a final prediction result;
and 43, writing the prediction result into a seismic channel corresponding to the sgy-format seismic data to generate a three-dimensional data body of the 'dessert' parameter, and analyzing the result by using professional seismic data visualization software.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111369371.3A CN114114414A (en) | 2021-11-18 | 2021-11-18 | Artificial intelligence prediction method for 'dessert' information of shale reservoir |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111369371.3A CN114114414A (en) | 2021-11-18 | 2021-11-18 | Artificial intelligence prediction method for 'dessert' information of shale reservoir |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114114414A true CN114114414A (en) | 2022-03-01 |
Family
ID=80397576
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111369371.3A Pending CN114114414A (en) | 2021-11-18 | 2021-11-18 | Artificial intelligence prediction method for 'dessert' information of shale reservoir |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114114414A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116168172A (en) * | 2023-04-19 | 2023-05-26 | 武汉中旺亿能科技发展有限公司 | Shale oil gas dessert prediction method, device, equipment and storage medium |
| CN117272841A (en) * | 2023-11-21 | 2023-12-22 | 西南石油大学 | Shale gas dessert prediction method based on hybrid neural network |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107870368A (en) * | 2016-09-26 | 2018-04-03 | 中国石油化工股份有限公司 | A kind of total content of organic carbon spatial distribution Forecasting Methodology based on seismic properties |
| CN108629072A (en) * | 2018-03-12 | 2018-10-09 | 山东科技大学 | Convolutional neural networks study towards the distribution of earthquake oil and gas reservoir and prediction technique |
| CN108897042A (en) * | 2018-08-28 | 2018-11-27 | 中国石油天然气股份有限公司 | Content of organic matter earthquake prediction method and device |
| CN109799533A (en) * | 2018-12-28 | 2019-05-24 | 中国石油化工股份有限公司 | A kind of method for predicting reservoir based on bidirectional circulating neural network |
| CN110954948A (en) * | 2018-09-27 | 2020-04-03 | 中国石油化工股份有限公司 | Physical property parameter inversion method and system for rock physical constraint reservoir |
| US20200211126A1 (en) * | 2018-12-29 | 2020-07-02 | Petrochina Company Limited | Prediction method for shale oil and gas sweet spot region, computer device and computer readable storage medium |
| CN111783825A (en) * | 2020-05-26 | 2020-10-16 | 中国石油天然气集团有限公司 | Well logging lithology identification method based on convolutional neural network learning |
| CN112017289A (en) * | 2020-08-31 | 2020-12-01 | 电子科技大学 | Well-seismic combined initial lithology model construction method based on deep learning |
| CN112505778A (en) * | 2020-12-01 | 2021-03-16 | 西南石油大学 | Three-dimensional in-situ characterization method for heterogeneity of shale storage and generation performance |
| CN112578475A (en) * | 2020-11-23 | 2021-03-30 | 中海石油(中国)有限公司 | Compact reservoir dual-dessert identification method based on data mining |
| US20210350208A1 (en) * | 2020-05-11 | 2021-11-11 | China University Of Petroleum (East China) | Method and device for predicting production performance of oil reservoir |
-
2021
- 2021-11-18 CN CN202111369371.3A patent/CN114114414A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107870368A (en) * | 2016-09-26 | 2018-04-03 | 中国石油化工股份有限公司 | A kind of total content of organic carbon spatial distribution Forecasting Methodology based on seismic properties |
| CN108629072A (en) * | 2018-03-12 | 2018-10-09 | 山东科技大学 | Convolutional neural networks study towards the distribution of earthquake oil and gas reservoir and prediction technique |
| CN108897042A (en) * | 2018-08-28 | 2018-11-27 | 中国石油天然气股份有限公司 | Content of organic matter earthquake prediction method and device |
| CN110954948A (en) * | 2018-09-27 | 2020-04-03 | 中国石油化工股份有限公司 | Physical property parameter inversion method and system for rock physical constraint reservoir |
| CN109799533A (en) * | 2018-12-28 | 2019-05-24 | 中国石油化工股份有限公司 | A kind of method for predicting reservoir based on bidirectional circulating neural network |
| US20200211126A1 (en) * | 2018-12-29 | 2020-07-02 | Petrochina Company Limited | Prediction method for shale oil and gas sweet spot region, computer device and computer readable storage medium |
| US20210350208A1 (en) * | 2020-05-11 | 2021-11-11 | China University Of Petroleum (East China) | Method and device for predicting production performance of oil reservoir |
| CN111783825A (en) * | 2020-05-26 | 2020-10-16 | 中国石油天然气集团有限公司 | Well logging lithology identification method based on convolutional neural network learning |
| CN112017289A (en) * | 2020-08-31 | 2020-12-01 | 电子科技大学 | Well-seismic combined initial lithology model construction method based on deep learning |
| CN112578475A (en) * | 2020-11-23 | 2021-03-30 | 中海石油(中国)有限公司 | Compact reservoir dual-dessert identification method based on data mining |
| CN112505778A (en) * | 2020-12-01 | 2021-03-16 | 西南石油大学 | Three-dimensional in-situ characterization method for heterogeneity of shale storage and generation performance |
Non-Patent Citations (1)
| Title |
|---|
| 段友祥: "卷积神经网络在储层预测中的应用研究", 《通信学报》, vol. 37, no. 1, pages 2 - 5 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116168172A (en) * | 2023-04-19 | 2023-05-26 | 武汉中旺亿能科技发展有限公司 | Shale oil gas dessert prediction method, device, equipment and storage medium |
| CN117272841A (en) * | 2023-11-21 | 2023-12-22 | 西南石油大学 | Shale gas dessert prediction method based on hybrid neural network |
| CN117272841B (en) * | 2023-11-21 | 2024-01-26 | 西南石油大学 | Shale gas dessert prediction method based on hybrid neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113553780B (en) | Stratum pore pressure prediction method based on machine learning | |
| CN109345007B (en) | Advantageous reservoir development area prediction method based on XGboost feature selection | |
| CN112444841B (en) | Thin-layer-containing lithology earthquake prediction method based on scale-division multi-input convolution network | |
| CN111596978A (en) | Web page display method, module and system for lithofacies classification by artificial intelligence | |
| CN112784980B (en) | Intelligent logging horizon dividing method | |
| CN113610945B (en) | Ground stress curve prediction method based on hybrid neural network | |
| CN114114414A (en) | Artificial intelligence prediction method for 'dessert' information of shale reservoir | |
| CN111665560A (en) | Oil-gas reservoir identification method and device, computer equipment and readable storage medium | |
| CN111027882A (en) | Method for evaluating brittleness index by utilizing conventional logging data based on high-order neural network | |
| CN114723095A (en) | Missing well logging curve prediction method and device | |
| CN112100930B (en) | Formation pore pressure calculation method based on convolutional neural network and Eaton formula | |
| CN112796738A (en) | Stratum permeability calculation method combining array acoustic logging and conventional logging | |
| CN114529110A (en) | Lithofacies inversion method and system based on deep neural network model | |
| CN110988997A (en) | Hydrocarbon source rock three-dimensional space distribution quantitative prediction technology based on machine learning | |
| CN112253087A (en) | Biological disturbance reservoir physical property calculation method based on multi-source logging data | |
| CN116305850A (en) | Stratum thermal conductivity prediction method based on random forest model | |
| CN107831540A (en) | The direct new method for extracting of reservoir physical parameter | |
| CN114065909A (en) | Well logging curve completion method based on CNN _ AB _ Bi-LSTM | |
| CN110118994B (en) | Continental facies hydrocarbon source rock quantitative prediction method based on seismic inversion and machine learning | |
| CN117251802A (en) | Heterogeneous reservoir parameter prediction method and system based on transfer learning | |
| CN109655394B (en) | Nuclear magnetic resonance T2 spectrum permeability calculation method under constraint of throat structure parameters | |
| CN111751878A (en) | Method and device for predicting transverse wave velocity | |
| CN114755744A (en) | Total organic carbon well logging interpretation method and system based on mud shale heterogeneity characteristics | |
| CN117407841B (en) | Shale layer seam prediction method based on optimization integration algorithm | |
| CN114075973B (en) | Stratum element logging curve reconstruction method and device |
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 |