CN116992754A - Rapid inversion method for logging while drilling data based on transfer learning - Google Patents
Rapid inversion method for logging while drilling data based on transfer learning Download PDFInfo
- Publication number
- CN116992754A CN116992754A CN202310750062.3A CN202310750062A CN116992754A CN 116992754 A CN116992754 A CN 116992754A CN 202310750062 A CN202310750062 A CN 202310750062A CN 116992754 A CN116992754 A CN 116992754A
- Authority
- CN
- China
- Prior art keywords
- logging
- drilling
- data
- electromagnetic wave
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000005553 drilling Methods 0.000 title claims abstract description 258
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000013526 transfer learning Methods 0.000 title claims abstract description 31
- 230000004044 response Effects 0.000 claims abstract description 97
- 238000012549 training Methods 0.000 claims abstract description 39
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 30
- 230000015572 biosynthetic process Effects 0.000 claims description 15
- 238000013528 artificial neural network Methods 0.000 claims description 14
- 238000010606 normalization Methods 0.000 claims description 14
- 238000004422 calculation algorithm Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 9
- 238000004088 simulation Methods 0.000 claims description 7
- 238000012795 verification Methods 0.000 claims description 7
- 238000013508 migration Methods 0.000 claims description 4
- 230000005012 migration Effects 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 4
- 230000001419 dependent effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- E—FIXED CONSTRUCTIONS
- E21—EARTH OR ROCK DRILLING; MINING
- E21B—EARTH OR ROCK DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS
- E21B47/00—Survey of boreholes or wells
- E21B47/12—Means for transmitting measuring-signals or control signals from the well to the surface, or from the surface to the well, e.g. for logging while drilling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/096—Transfer learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Mining & Mineral Resources (AREA)
- Geology (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Probability & Statistics with Applications (AREA)
- Remote Sensing (AREA)
- Geophysics (AREA)
- Environmental & Geological Engineering (AREA)
- Fluid Mechanics (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Geochemistry & Mineralogy (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention discloses a fast inversion method of logging while drilling data based on transfer learning, which comprises the steps of constructing a forward model of logging while drilling azimuth electromagnetic wave, simulating logging while drilling response of a multi-component logging while drilling electromagnetic wave logging instrument under different stratum parameter conditions by using the forward model of logging while drilling azimuth electromagnetic wave, constructing a logging while drilling response database, constructing an inversion model of logging while drilling azimuth electromagnetic wave logging data by combining a convolutional neural network and an LSTM network, training and optimizing logging while drilling response data in the logging while drilling response database to obtain a source domain network model, constructing a target domain network model based on transfer learning, obtaining an inversion model of logging while drilling data based on transfer learning after training the target domain network model by using actual logging data, and verifying the accuracy of inversion results of the logging while drilling data inversion model.
Description
Technical Field
The invention relates to the technical field of oil and gas exploration and development, in particular to a method for quickly inverting logging while drilling data based on transfer learning.
Background
With the development of petroleum industry, small oil layers and thin interbings are becoming the main battlefield for oil and gas exploration and development. The geosteering while drilling technique can increase the drilling rate of the oilfield at the desired layer by optimizing the trajectory of the wellbore within the reservoir. In order to effectively achieve geosteering while drilling, well logging data in a reservoir is evaluated, and an effective inversion model needs to be built. The inversion model relates to a complex mapping relation between specific physical quantities, and stratum parameters are reconstructed from observed data to serve as a complex nonlinear function comprehensive optimizing process, so that the response rule is complex, nonlinear characteristics are strong, and the inversion of logging data is difficult to be performed quickly and accurately.
At present, the conventional inversion method mostly adopts a gradient algorithm and a random algorithm. When the gradient algorithm is adopted for inversion, a cost function for measuring the error of the true value and the predicted value is required to be constructed, an initial value is set, and the cost function is reduced through iterative calculation, so that an inversion result is obtained. However, in the iterative calculation process, the forward model needs to be continuously called and the Jacobian matrix needs to be repeatedly calculated, so that the time consumption is long, the inversion result is excessively dependent on the selection of the initial value and is easy to fall into a local minimum value, and the inversion result error is large; when the random algorithm is adopted for inversion, forward modeling needs to be repeated for a plurality of times at each logging position, which is very time-consuming, so that the random algorithm has larger limitation when being applied to logging data inversion, and is difficult to meet the requirement of on-site while-drilling operation. Therefore, how to implement accurate and real-time inversion of instruments in complex geological environments has become a problem to be solved in geosteering logging while drilling.
The rapid development of the deep learning method provides a new idea for solving the difficult problem in the field of earth exploration. Although the deep learning method can fit a strong nonlinear mapping relation and meets the requirement of real-time inversion of logging while drilling data in geosteering while drilling, the accuracy of the deep learning method is extremely dependent on the complexity of a network structure and the sample size of training data, and the application of the deep learning in the geophysical field is seriously hindered by the problem of small sample size of labels. Therefore, it is needed to provide a logging while drilling data inversion method based on transfer learning, which overcomes the problem of less logging while drilling measured data in the inversion process.
Disclosure of Invention
The invention aims to provide a fast inversion method of logging while drilling data based on transfer learning, which can realize fast inversion of logging data based on a small amount of actual logging while drilling data, improves inversion accuracy of an inversion model of logging while drilling data, solves the problems that logging while drilling data inversion excessively depends on an initial value and consumes longer time, and has good application prospect.
In order to achieve the above purpose, the invention adopts the following technical scheme:
a migration learning-based quick inversion method for logging while drilling data comprises the following steps:
step 1, establishing a forward model of the electromagnetic wave logging while drilling azimuth according to instrument structures of the electromagnetic wave logging while drilling instrument with multiple components and stratum parameters of a layered anisotropic stratum, wherein the forward model of the electromagnetic wave logging while drilling azimuth comprises the layered anisotropic stratum model and the electromagnetic wave logging while drilling instrument model with multiple components, and simulating by using the forward model of the electromagnetic wave logging while drilling azimuth to obtain logging while drilling response of the electromagnetic wave logging while drilling instrument with multiple components in the layered anisotropic stratum;
step 2, setting the frequency and source distance of a transmitting coil in a multi-component while-drilling electromagnetic wave logging instrument model in a while-drilling azimuth electromagnetic wave logging forward model, sequentially changing the stratum inclination angle theta, the stratum horizontal resistivity Rh, the stratum vertical resistivity Rv and the anisotropy coefficient lambda of a layer-shaped anisotropic stratum model in the while-drilling azimuth electromagnetic wave logging forward model, simulating to obtain a while-drilling logging response curve of the multi-component while-drilling electromagnetic wave logging instrument under different stratum parameter conditions, and analyzing the change rule of the while-drilling logging response curve along with each stratum parameter;
step 3, according to logging-while-drilling response curves obtained by simulation of forward modeling of electromagnetic wave logging while drilling in different stratum parameter conditions, acquiring logging-while-drilling responses of the multi-component electromagnetic wave logging while drilling instrument under different stratum parameter conditions, using the logging-while-drilling responses of the multi-component electromagnetic wave logging while drilling instrument as source domain data, using stratum parameters corresponding to the logging-while-drilling responses as labels, generating logging-while-drilling response data, and constructing a logging-while-drilling response database;
step 4, preprocessing each logging-while-drilling response data in the logging-while-drilling response database, normalizing each logging-while-drilling response data in the logging-while-drilling response database to obtain normalized logging-while-drilling response data, and dividing the normalized logging-while-drilling response data in the logging-while-drilling response database to obtain a training set and a verification set;
step 5, constructing a logging-while-drilling azimuth electromagnetic wave logging data inversion model based on a neural network combining a convolutional neural network and an LSTM network, wherein the input of the logging-while-drilling azimuth electromagnetic wave logging data inversion model is a logging-while-drilling response curve, and the output of the logging-while-drilling azimuth electromagnetic wave logging data inversion model is stratum parameters;
step 6, sequentially inputting logging-while-drilling response data in a training set into an inversion model of logging-while-drilling azimuth electromagnetic wave logging data, training the inversion model of logging-while-drilling azimuth electromagnetic wave logging data to obtain stratum parameters according to inversion of logging-while-drilling response, and training the inversion model of logging-while-drilling azimuth electromagnetic wave logging data until a loss function value of the inversion model of logging-while-drilling azimuth electromagnetic wave logging data is smaller than a preset precision value, thereby obtaining the inversion model of logging-while-drilling azimuth electromagnetic wave logging data after training;
step 7, verifying the trained logging-while-drilling azimuth electromagnetic wave logging data inversion model by using the verification concentrated logging-while-drilling response data, inverting stratum parameters by using the trained logging-while-drilling azimuth electromagnetic wave logging data inversion model, determining a loss function value of the logging-while-drilling azimuth electromagnetic wave logging data inversion model in the inversion process, outputting the trained logging-while-drilling azimuth electromagnetic wave logging data inversion model as a source domain network model if the loss function value of the logging-while-drilling azimuth electromagnetic wave logging data inversion model is smaller than a preset precision value, otherwise, adjusting super parameters of a neural network, and returning to the step 6 to continue training the logging-while-drilling azimuth electromagnetic wave logging data inversion model;
step 8, based on transfer learning, constructing a target domain network model by using a source domain network model, acquiring actual logging data, carrying out normalization processing on the actual logging data, then constructing an actual logging data set by using the actual logging data as target threshold data, randomly initializing weight parameters of the last layer in a neural network for constructing the source domain network model, sequentially training the target domain network model by using each target threshold data in the actual logging data set, transferring the source domain network model to an actual logging data inversion process, acquiring a loss function value of the target domain network model, adjusting super parameters of the neural network in the target domain network model if the loss function value is smaller than a preset precision value, and continuously training the target domain network model by using the target threshold data, otherwise, outputting the trained target domain network model to obtain the inversion model of the logging-while-drilling data based on transfer learning;
and 9, outputting a logging while drilling data inversion model based on transfer learning, and verifying the accuracy of inversion results of the logging while drilling data inversion model based on transfer learning.
Preferably, in the azimuth while drilling electromagnetic wave logging forward model, the multi-component electromagnetic wave logging instrument model while drilling is positioned in a layered anisotropic stratum model, and the layered anisotropic stratum model is arranged according to stratum parameters of a layered anisotropic stratum, and a plurality of stratum interfaces are sequentially arranged from top to bottom; the formation parameters of the layered anisotropic formation include azimuth angleFormation dip θ, formation horizontal resistivity Rh, formation vertical resistivity Rv, and anisotropy coefficient λ;
the multi-component electromagnetic wave logging instrument while drilling model is internally provided with a transmitting coil system and a receiving coil system, the distance between the transmitting coil system and the receiving coil system is the source distance, three transmitting coils in the transmitting coil system are mutually orthogonal, three receiving coils in the receiving coil system are mutually orthogonal, the frequency and the source distance of the transmitting coils in the multi-component electromagnetic wave logging instrument while drilling model are set, and the logging while drilling response of the multi-component electromagnetic wave logging instrument while drilling in the layered anisotropic stratum is obtained by utilizing the forward modeling simulation of the electromagnetic wave logging while drilling model.
Preferably, in the step 4, the calculation formula of normalization processing of logging while drilling response data is:
wherein x' is logging-while-drilling response data after normalization processing, x is logging-while-drilling response data before processing, mu is the mean value of logging-while-drilling response, and delta is the standard deviation of logging-while-drilling response;
preferably, the LSTM network has three inputs at time t: cell state C t-1 Hidden layer state h t-1 And the current input feature x t ；
In the LSTM network, a forget door f is arranged in a unit door t Update gate i t And an output gate o t ；
Cell door state C t The method comprises the following steps:
wherein,,
f t ＝σ(W f ·[h t-1 ,x t ]+b f ) (3)
i t ＝σ(W i ·[h t-1 ,x t ]+b i ) (4)
wherein f t For forgetting the door, control the discard unit state C t-1 Certain information of W f Weight of forgetting gate b f Bias for forgetting the door; x is x t Is the input variable at the time t, h t-1 The hidden layer state at the time t-1; i.e t To update the gate, select to preserve candidate cell statesCertain information of W C To update the weight of the door b C To update the bias of the gate; />Updating values for cell state, W C Weights for new candidate cell states, b C Bias for new candidate cell states; sigma is a softmax function for compressing data to a range of 0-1;
hidden layer output result h t From the output gate o t And the unit door state C calculated in the last step t Calculated by the tanh activation function, as shown in formula (6):
h t ＝o t *tanh(C t ) (6)
wherein,,
o t ＝σ(W o [h t-1 ,x t ]+b o ) (7)
in the formula, o t For the output door, W o To output the weight of the door, b o To output the gate bias.
Preferably, training an inversion model of the electromagnetic wave logging data of the azimuth while drilling based on Batch Normalization algorithm is carried out as shown in a formula (8):
wherein x= (x) 1 ,x 2 ,…,x k ,…,x d ) For input data, the dimension is d;gamma is the data processed by Batch Normalization algorithm k And beta k Are all learning parameters, gamma k The initial value of (1) < beta- k The initial value of (a) is set to 0, E is the mathematical expectation of the input data, var is the standard deviation of the input data, epsilon is the minimum number larger than zero;
the loss function value of the inversion model of the azimuth electromagnetic wave logging while drilling data is as follows:
in the loss of MSE For the loss function value, i is y i For the true value of the formation parameter corresponding to the ith logging while drilling response data in the training set,and adopting the calculation result of the inversion model of the electromagnetic wave logging data of the azimuth while drilling for the ith logging while drilling response data in the training set, wherein n is the total number of logging while drilling response data in the training set.
Preferably, the precision value is set to 5%.
The invention has the following advantages:
according to the invention, the convolution neural network and the LSTM network are combined to construct the inversion model of the azimuth electromagnetic wave logging while drilling, and the migration algorithm is combined, so that the data volume of sample data required by the training process of the inversion model of the azimuth electromagnetic wave logging while drilling is reduced, the calculation time is greatly shortened by optimizing the neural network structure, the inversion precision and the inversion speed of the inversion model of the azimuth electromagnetic wave logging while drilling are improved, the problem of gradient hours in the inversion process is solved, and the inversion model of the azimuth electromagnetic wave logging while drilling based on migration learning realizes accurate inversion of geological parameters according to the response of logging while drilling under the condition of less actual logging while drilling data, and has good application prospect.
Drawings
FIG. 1 is a flow chart of a method for fast inversion of logging while drilling data based on transfer learning.
FIG. 2 is a schematic structural diagram of a forward model of electromagnetic wave logging while drilling azimuth; t in FIG. 2 x T is the x-axis direction of the transmitting coil system y For the y-axis direction of the transmit coil system, T z R is the z-axis direction of the transmitting coil system x For receiving the x-axis direction of the coil system, R y For receiving the y-axis direction of the coil system, R z Is the z-axis direction of the receive coil system.
FIG. 3 shows apparent conductivity components obtained by simulation of an electromagnetic wave logging forward model of azimuth while drilling under different stratum dip angles.
Fig. 4 is a schematic diagram of the LSTM network.
Detailed Description
The invention is described in further detail below with reference to the attached drawings and detailed description:
the embodiment provides a fast inversion method of logging while drilling data based on transfer learning, as shown in fig. 1, specifically comprising the following steps:
step 1, acquiring instrument structures of a multi-component electromagnetic wave logging instrument while drilling and stratum parameters of a layered anisotropic stratum, wherein the stratum parameters comprise azimuth anglesAnd (3) establishing a forward model of electromagnetic wave logging while drilling azimuth according to the instrument structure of the multi-component electromagnetic wave logging while drilling instrument and the stratum parameters of the layered anisotropic stratum.
The forward model of the electromagnetic wave logging while drilling azimuth is shown in fig. 2, and comprises a layered anisotropic stratum model and a multi-component electromagnetic wave logging while drilling instrument model, wherein a transmitting coil system and a receiving coil system are arranged in the multi-component electromagnetic wave logging while drilling instrument model, the distance between the transmitting coil system and the receiving coil system is a source distance, three transmitting coils in the transmitting coil system are mutually orthogonal, three receiving coils in the receiving coil system are mutually orthogonal, and the frequency and the source distance of the transmitting coils in the multi-component electromagnetic wave logging while drilling instrument model are set.
The well logging response of the multi-component electromagnetic logging while drilling instrument in the layered anisotropic stratum is obtained by setting the transmitting frequency and source distance of a transmitting coil system in the multi-component electromagnetic logging while drilling instrument model and utilizing the forward model of the electromagnetic logging while drilling azimuth.
And 2, setting the frequency and source distance of a transmitting coil in a multi-component electromagnetic wave logging instrument model while drilling in the electromagnetic wave logging forward model while drilling, sequentially changing the stratum inclination angle theta, the stratum horizontal resistivity Rh, the stratum vertical resistivity Rv and the anisotropy coefficient lambda of a layer-shaped anisotropic stratum model in the electromagnetic wave logging forward model while drilling, and obtaining the logging-while-drilling response curve of the multi-component electromagnetic wave logging instrument while drilling under different stratum parameter conditions by simulation, and analyzing the change rule of the logging-while-drilling response curve with each stratum parameter.
In this embodiment, when the frequency and the source distance of the transmitting coil in the multi-component while-drilling electromagnetic wave logging instrument model are fixed in the while-drilling azimuth electromagnetic wave logging forward model, and the formation dip angle θ of the layered anisotropic formation model is set to 60 °, 70 °, 80 ° and 90 ° in sequence, the while-drilling logging response of the multi-component while-drilling electromagnetic wave logging instrument in the layered anisotropic formation under different formation dip angles is simulated by using the while-drilling azimuth electromagnetic wave logging forward model, as shown in fig. 3.
And 3, according to a logging-while-drilling response curve obtained by simulation of the forward model of the electromagnetic wave logging while drilling in azimuth under different stratum parameter conditions, acquiring logging-while-drilling response of the multi-component electromagnetic wave logging while drilling instrument under different stratum parameter conditions, taking the logging-while-drilling response of the multi-component electromagnetic wave logging while drilling instrument as source domain data, taking stratum parameters corresponding to the logging-while-drilling response as labels, generating logging-while-drilling response data, and constructing a logging-while-drilling response database.
And 4, preprocessing each piece of logging-while-drilling response data in the logging-while-drilling response database, wherein the preprocessing comprises supplementing missing values, denoising and outlier rejection, normalizing each piece of logging-while-drilling response data in the logging-while-drilling response database by utilizing a logging-while-drilling response data normalization processing calculation formula to obtain normalized logging-while-drilling response data, normalizing the logging-while-drilling response data in the logging-while-drilling response database to the same scale, and dividing the normalized logging-while-drilling response data in the logging-while-drilling response database to obtain a training set and a verification set.
In this embodiment, the calculation formula of normalization processing of logging while drilling response data is as follows:
where x' is normalized logging-while-drilling response data, x is pre-processed logging-while-drilling response data, μ is the mean value of the logging-while-drilling response, and δ is the standard deviation of the logging-while-drilling response.
And 5, constructing a logging-while-drilling azimuth electromagnetic wave logging data inversion model based on a neural network combining the convolutional neural network and the LSTM network, wherein the input of the logging-while-drilling azimuth electromagnetic wave logging data inversion model is a logging-while-drilling response curve, the output of the logging-while-drilling azimuth electromagnetic wave logging data inversion model is stratum parameters, and an Adam optimizer algorithm is adopted.
In this embodiment, as shown in fig. 4, the LSTM network has three inputs at time t: cell state C t-1 Hidden layer state h t-1 And the current input feature x t 。
In the LSTM network, a forget door f is arranged in a unit door t Update gate i t And an output gate o t ；
Cell door state C t The method comprises the following steps:
wherein,,
f t ＝σ(W f ·[h t-1 ,x t ]+b f ) (3)
i t ＝σ(W i ·[h t-1 ,x t ]+b i ) (4)
wherein f t For forgetting the door, control the discard unit state C t-1 Certain information of W f Weight of forgetting gate b f Bias for forgetting the door; x is x t Is the input variable at the time t, h t-1 The hidden layer state at the time t-1; i.e t To update the gate, select to preserve candidate cell statesCertain information of W C To update the weight of the door b C To update the bias of the gate; />Updating values for cell state, W C Weights for new candidate cell states, b C Bias for new candidate cell states; sigma is a softmax function for compressing data to a range of 0-1;
hidden layer output result h t From the output gate o t And the unit door state C calculated in the last step t Calculated by the tanh activation function, as shown in formula (6):
h t ＝o t *tanh(C t ) (6)
wherein,,
o t ＝σ(W o [h t-1 ,x t ]+b o ) (7)
in the formula, o t For the output door, W o To output the weight of the door, b o To output the gate bias.
And 6, sequentially inputting logging-while-drilling response data in a training set into the logging-while-drilling azimuth electromagnetic wave logging information inversion model, training the logging-while-drilling azimuth electromagnetic wave logging information inversion model to obtain stratum parameters according to logging-while-drilling response inversion, and training the logging-while-drilling azimuth electromagnetic wave logging information inversion model until the loss function value of the logging-while-drilling azimuth electromagnetic wave logging information inversion model is smaller than a preset precision value by 5%, thereby obtaining the trained logging-while-drilling azimuth electromagnetic wave logging information inversion model.
In order to prevent the problem of gradient disappearance in the training process, the embodiment trains the inversion model of the electromagnetic wave logging data of the azimuth while drilling based on Batch Normalization algorithm, as shown in formula (8):
wherein x= (x) 1 ,x 2 ,…,x k ,…,x d ) For input data, the dimension is d;gamma is the data processed by Batch Normalization algorithm k And beta k Are all learning parameters, gamma k The initial value of (1) < beta- k The initial value is set to 0, E is the mathematical expectation of the input data, and the learning parameter gamma in the training process k And beta k Continuously updating along with the training times; var is the standard deviation of the input data and ε is a fractional number greater than zero.
In this embodiment, the loss function value of the inversion model of the logging-while-drilling azimuth electromagnetic wave logging data is:
in the loss of MSE For the loss function value, i is y i The ith while-drilling in the training setThe true values of the formation parameters corresponding to the log response data,and adopting the calculation result of the inversion model of the electromagnetic wave logging data of the azimuth while drilling for the ith logging while drilling response data in the training set, wherein n is the total number of logging while drilling response data in the training set.
And 7, verifying the trained logging-while-drilling azimuth electromagnetic wave logging data inversion model by using the verification concentrated logging-while-drilling response data, inverting stratum parameters by using the trained logging-while-drilling azimuth electromagnetic wave logging data inversion model, determining a loss function value of the logging-while-drilling azimuth electromagnetic wave logging data inversion model in the inversion process, outputting the trained logging-while-drilling azimuth electromagnetic wave logging data inversion model as a source domain network model if the loss function value of the logging-while-drilling azimuth electromagnetic wave logging data inversion model is smaller than a preset precision value by 5%, otherwise, adjusting super parameters of a neural network, and returning to the step 6 to continuously train the logging-while-drilling azimuth electromagnetic wave logging data inversion model.
And 8, constructing a target domain network model by using a source domain network model based on transfer learning, acquiring actual logging data, carrying out normalization processing on the actual logging data, constructing an actual logging data set by using the actual logging data as target threshold data, randomly initializing weight parameters of the last layer in a neural network for constructing the source domain network model, sequentially training the target domain network model by using each target threshold data in the actual logging data set, transferring the source domain network model to an actual logging data inversion process, acquiring a loss function value of the target domain network model, adjusting super parameters of the neural network in the target domain network model if the loss function value is smaller than a preset precision value by 5%, and continuously training the target domain network model by using the target threshold data, otherwise, outputting the trained target domain network model, and obtaining the well logging while drilling data inversion model based on transfer learning.
And 9, outputting a logging while drilling data inversion model based on transfer learning, verifying the accuracy of inversion results of the logging while drilling data inversion model based on transfer learning, and inputting the logging while drilling data inversion model based on transfer learning by using logging while drilling response data in verification set.
In the embodiment, inversion is performed by using a logging while drilling data inversion model based on transfer learning, the accuracy, inversion speed and generalization capability of an inversion result are obtained, and after the accuracy, inversion speed and generalization capability of the inversion result are determined to meet the requirements, the logging while drilling data inversion model based on transfer learning is packaged, so that the logging while drilling data inversion model meeting the logging while drilling data inversion requirements is obtained based on transfer learning.
The foregoing description is, of course, merely illustrative of preferred embodiments of the present invention, and it should be understood that the present invention is not limited to the above-described embodiments, but is intended to cover all modifications, equivalents and alternatives falling within the spirit and scope of the present invention as defined by the appended claims.
Claims (6)
1. A migration learning-based quick inversion method for logging-while-drilling data is characterized by comprising the following steps:
step 1, establishing a forward model of the electromagnetic wave logging while drilling azimuth according to instrument structures of the electromagnetic wave logging while drilling instrument with multiple components and stratum parameters of a layered anisotropic stratum, wherein the forward model of the electromagnetic wave logging while drilling azimuth comprises the layered anisotropic stratum model and the electromagnetic wave logging while drilling instrument model with multiple components, and simulating by using the forward model of the electromagnetic wave logging while drilling azimuth to obtain logging while drilling response of the electromagnetic wave logging while drilling instrument with multiple components in the layered anisotropic stratum;
step 2, setting the frequency and source distance of a transmitting coil in a multi-component while-drilling electromagnetic wave logging instrument model in a while-drilling azimuth electromagnetic wave logging forward model, sequentially changing the stratum inclination angle theta, the stratum horizontal resistivity Rh, the stratum vertical resistivity Rv and the anisotropy coefficient lambda of a layer-shaped anisotropic stratum model in the while-drilling azimuth electromagnetic wave logging forward model, simulating to obtain a while-drilling logging response curve of the multi-component while-drilling electromagnetic wave logging instrument under different stratum parameter conditions, and analyzing the change rule of the while-drilling logging response curve along with each stratum parameter;
step 3, according to logging-while-drilling response curves obtained by simulation of forward modeling of electromagnetic wave logging while drilling in different stratum parameter conditions, acquiring logging-while-drilling responses of the multi-component electromagnetic wave logging while drilling instrument under different stratum parameter conditions, using the logging-while-drilling responses of the multi-component electromagnetic wave logging while drilling instrument as source domain data, using stratum parameters corresponding to the logging-while-drilling responses as labels, generating logging-while-drilling response data, and constructing a logging-while-drilling response database;
step 4, preprocessing each logging-while-drilling response data in the logging-while-drilling response database, normalizing each logging-while-drilling response data in the logging-while-drilling response database to obtain normalized logging-while-drilling response data, and dividing the normalized logging-while-drilling response data in the logging-while-drilling response database to obtain a training set and a verification set;
step 5, constructing a logging-while-drilling azimuth electromagnetic wave logging data inversion model based on a neural network combining a convolutional neural network and an LSTM network, wherein the input of the logging-while-drilling azimuth electromagnetic wave logging data inversion model is a logging-while-drilling response curve, and the output of the logging-while-drilling azimuth electromagnetic wave logging data inversion model is stratum parameters;
step 6, sequentially inputting logging-while-drilling response data in a training set into an inversion model of logging-while-drilling azimuth electromagnetic wave logging data, training the inversion model of logging-while-drilling azimuth electromagnetic wave logging data to obtain stratum parameters according to inversion of logging-while-drilling response, and training the inversion model of logging-while-drilling azimuth electromagnetic wave logging data until a loss function value of the inversion model of logging-while-drilling azimuth electromagnetic wave logging data is smaller than a preset precision value, thereby obtaining the inversion model of logging-while-drilling azimuth electromagnetic wave logging data after training;
step 7, verifying the trained logging-while-drilling azimuth electromagnetic wave logging data inversion model by using the verification concentrated logging-while-drilling response data, inverting stratum parameters by using the trained logging-while-drilling azimuth electromagnetic wave logging data inversion model, determining a loss function value of the logging-while-drilling azimuth electromagnetic wave logging data inversion model in the inversion process, outputting the trained logging-while-drilling azimuth electromagnetic wave logging data inversion model as a source domain network model if the loss function value of the logging-while-drilling azimuth electromagnetic wave logging data inversion model is smaller than a preset precision value, otherwise, adjusting super parameters of a neural network, and returning to the step 6 to continue training the logging-while-drilling azimuth electromagnetic wave logging data inversion model;
step 8, based on transfer learning, constructing a target domain network model by using a source domain network model, acquiring actual logging data, carrying out normalization processing on the actual logging data, then constructing an actual logging data set by using the actual logging data as target threshold data, randomly initializing weight parameters of the last layer in a neural network for constructing the source domain network model, sequentially training the target domain network model by using each target threshold data in the actual logging data set, transferring the source domain network model to an actual logging data inversion process, acquiring a loss function value of the target domain network model, adjusting super parameters of the neural network in the target domain network model if the loss function value is smaller than a preset precision value, and continuously training the target domain network model by using the target threshold data, otherwise, outputting the trained target domain network model to obtain the inversion model of the logging-while-drilling data based on transfer learning;
and 9, outputting a logging while drilling data inversion model based on transfer learning, and verifying the accuracy of inversion results of the logging while drilling data inversion model based on transfer learning.
2. The rapid inversion method of logging while drilling data based on transfer learning of claim 1, wherein in the forward model of electromagnetic wave logging while drilling azimuth, a multi-component electromagnetic wave logging while drilling instrument model is positioned in a layered anisotropic stratum model, and the layered anisotropic stratum model is arranged according to stratum parameters of a layered anisotropic stratum, and a plurality of stratum interfaces are sequentially arranged from top to bottom; the formation parameters of the layered anisotropic formation include azimuth angleFormation dip θ, formation horizontal resistivity Rh, formation vertical resistivity Rv, and anisotropy coefficient λ;
the multi-component electromagnetic wave logging instrument while drilling model is internally provided with a transmitting coil system and a receiving coil system, the distance between the transmitting coil system and the receiving coil system is the source distance, three transmitting coils in the transmitting coil system are mutually orthogonal, three receiving coils in the receiving coil system are mutually orthogonal, the frequency and the source distance of the transmitting coils in the multi-component electromagnetic wave logging instrument while drilling model are set, and the logging while drilling response of the multi-component electromagnetic wave logging instrument while drilling in the layered anisotropic stratum is obtained by utilizing the forward modeling simulation of the electromagnetic wave logging while drilling model.
3. The fast inversion method of logging while drilling data based on transfer learning of claim 1, wherein in the step 4, a calculation formula of normalization processing of logging while drilling response data is:
where x' is normalized logging-while-drilling response data, x is pre-processed logging-while-drilling response data, μ is the mean value of the logging-while-drilling response, and δ is the standard deviation of the logging-while-drilling response.
4. The rapid inversion method of logging while drilling data based on transfer learning of claim 1, wherein the LSTM network has three inputs at time t: cell state C t-1 Hidden layer state h t-1 And the current input feature x t ；
In the LSTM network, a forget door f is arranged in a unit door t Update gate i t And an output gate o t ；
Cell door state C t The method comprises the following steps:
wherein,,
f t ＝σ(W f ·[h t-1 ,x t ]+b f ) (3)
i t ＝σ(W i ·[h t-1 ,x t ]+b i ) (4)
wherein f t For forgetting the door, control the discard unit state C t-1 Certain information of W f Weight of forgetting gate b f Bias for forgetting the door; x is x t Is the input variable at the time t, h t-1 The hidden layer state at the time t-1; i.e t To update the gate, select to preserve candidate cell statesCertain information of W C To update the weight of the door b C To update the bias of the gate; />Updating values for cell state, W C Weights for new candidate cell states, b C Bias for new candidate cell states; sigma is a softmax function for compressing data to a range of 0-1;
hidden layer output result h t From the output gate o t And the unit door state C calculated in the last step t Calculated by the tanh activation function, as shown in formula (6):
h t ＝o t *tanh(C t ) (6)
wherein,,
o t ＝σ(W o [h t-1 ,x t ]+b o ) (7)
in the formula, o t For the output door, W o To output the weight of the door, b o To output the gate bias.
5. The rapid inversion method of logging while drilling data based on transfer learning of claim 4, wherein the inversion model of the electromagnetic wave logging while drilling azimuth is trained based on Batch Normalization algorithm, as shown in formula (8):
wherein x= (x) 1 ,x 2 ,…,x k ,…,x d ) For input data, the dimension is d;gamma is the data processed by Batch Normalization algorithm k And beta k Are all learning parameters, gamma k The initial value of (1) < beta- k The initial value of (a) is set to 0, E is the mathematical expectation of the input data, var is the standard deviation of the input data, epsilon is the minimum number larger than zero;
the loss function value of the inversion model of the azimuth electromagnetic wave logging while drilling data is as follows:
in the loss of MSE For the loss function value, i is y i For the true value of the formation parameter corresponding to the ith logging while drilling response data in the training set,and adopting the calculation result of the inversion model of the electromagnetic wave logging data of the azimuth while drilling for the ith logging while drilling response data in the training set, wherein n is the total number of logging while drilling response data in the training set.
6. The rapid inversion method of logging while drilling data based on transfer learning of claim 1, wherein the precision value is set to 5%.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310750062.3A CN116992754A (en) | 2023-06-25 | 2023-06-25 | Rapid inversion method for logging while drilling data based on transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310750062.3A CN116992754A (en) | 2023-06-25 | 2023-06-25 | Rapid inversion method for logging while drilling data based on transfer learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116992754A true CN116992754A (en) | 2023-11-03 |
Family
ID=88520319
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310750062.3A Pending CN116992754A (en) | 2023-06-25 | 2023-06-25 | Rapid inversion method for logging while drilling data based on transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116992754A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117744505A (en) * | 2024-02-20 | 2024-03-22 | 电子科技大学 | Deep learning-based inversion method for electromagnetic wave resistivity of azimuth while drilling |
| CN118171582A (en) * | 2024-05-11 | 2024-06-11 | 中国石油大学(华东) | Electromagnetic wave logging inversion method and system for azimuth while drilling by combining residual neural network and L-M algorithm |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108345719A (en) * | 2018-01-17 | 2018-07-31 | 中国石油大学(华东) | A kind of high angle hole/horizontal well is with boring inversion method after electromagnetic wave logging data is bored |
| CN115393656A (en) * | 2022-10-26 | 2022-11-25 | 中石化经纬有限公司 | Automatic classification method for stratum classification of logging-while-drilling image |
| CN115629428A (en) * | 2022-09-23 | 2023-01-20 | 中国海洋石油集团有限公司 | Method for performing joint inversion on reservoir parameters through resistivity and nuclear logging under slurry invasion |
| US20230074074A1 (en) * | 2021-09-02 | 2023-03-09 | Southwest Petroleum University | Intelligent recognition method for while-drilling safety risk based on convolutional neural network |
| CN115795994A (en) * | 2022-09-29 | 2023-03-14 | 西安石油大学 | Orientation electromagnetic wave logging while drilling data inversion method based on Unet convolution neural network |
-
2023
- 2023-06-25 CN CN202310750062.3A patent/CN116992754A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108345719A (en) * | 2018-01-17 | 2018-07-31 | 中国石油大学(华东) | A kind of high angle hole/horizontal well is with boring inversion method after electromagnetic wave logging data is bored |
| US20230074074A1 (en) * | 2021-09-02 | 2023-03-09 | Southwest Petroleum University | Intelligent recognition method for while-drilling safety risk based on convolutional neural network |
| CN115629428A (en) * | 2022-09-23 | 2023-01-20 | 中国海洋石油集团有限公司 | Method for performing joint inversion on reservoir parameters through resistivity and nuclear logging under slurry invasion |
| CN115795994A (en) * | 2022-09-29 | 2023-03-14 | 西安石油大学 | Orientation electromagnetic wave logging while drilling data inversion method based on Unet convolution neural network |
| CN115393656A (en) * | 2022-10-26 | 2022-11-25 | 中石化经纬有限公司 | Automatic classification method for stratum classification of logging-while-drilling image |
Non-Patent Citations (2)
| Title |
|---|
| GAOYANG ZHU 等: "A Fast Inversion of Induction Logging Data In Anisotropic Formation Based on Deep Learning", IEEE GEOSCIENCE AND REMOTE SENSING SOCIETY, 7 January 2020 (2020-01-07), pages 2050 - 2054, XP011822529, DOI: 10.1109/LGRS.2019.2961374 * |
| 蔡军;张恒荣;曾少军;高永德;程远方;: "随钻电磁波电阻率测井联合反演方法及其应用", 石油学报, no. 03, 15 March 2016 (2016-03-15), pages 371 - 381 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117744505A (en) * | 2024-02-20 | 2024-03-22 | 电子科技大学 | Deep learning-based inversion method for electromagnetic wave resistivity of azimuth while drilling |
| CN117744505B (en) * | 2024-02-20 | 2024-04-26 | 电子科技大学 | Depth learning-based inversion method for electromagnetic wave resistivity of azimuth while drilling |
| CN118171582A (en) * | 2024-05-11 | 2024-06-11 | 中国石油大学(华东) | Electromagnetic wave logging inversion method and system for azimuth while drilling by combining residual neural network and L-M algorithm |
| CN118171582B (en) * | 2024-05-11 | 2024-07-12 | 中国石油大学(华东) | Electromagnetic wave logging inversion method and system for azimuth while drilling by combining residual neural network and L-M algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116992754A (en) | Rapid inversion method for logging while drilling data based on transfer learning | |
| CN113486591B (en) | Gravity multi-parameter data density weighted inversion method for convolutional neural network result | |
| CN113610945B (en) | Ground stress curve prediction method based on hybrid neural network | |
| CN103454677B (en) | Based on the earthquake data inversion method that population is combined with linear adder device | |
| Chen et al. | Integration of principal-component-analysis and streamline information for the history matching of channelized reservoirs | |
| CN112733449A (en) | CNN well-seismic joint inversion method, CNN well-seismic joint inversion system, CNN well-seismic joint inversion storage medium, CNN well-seismic joint inversion equipment and CNN well-seismic joint inversion application | |
| Zhu et al. | Inversion of the permeability of a tight gas reservoir with the combination of a deep Boltzmann kernel extreme learning machine and nuclear magnetic resonance logging transverse relaxation time spectrum data | |
| Zhu et al. | A fast inversion of induction logging data in anisotropic formation based on deep learning | |
| CN117744505B (en) | Depth learning-based inversion method for electromagnetic wave resistivity of azimuth while drilling | |
| CN107256316B (en) | Artificial intelligence electromagnetic logging inversion method based on high-speed forward result training | |
| CN113204054A (en) | Self-adaptive wide-area electromagnetic method induced polarization information extraction method based on reinforcement learning | |
| Zhou et al. | Determination of pore size distribution in tight gas sandstones based on Bayesian regularization neural network with MICP, NMR and petrophysical logs | |
| Razak et al. | Embedding physical flow functions into deep learning predictive models for improved production forecasting | |
| CN106959121B (en) | Application method of self-contained reverse-optimization-based overrun learning algorithm in magnetic compass error compensation | |
| Hao et al. | A robust inversion of induction logging responses in anisotropic formation based on supervised descent method | |
| Wang et al. | On the Feasibility of An Ensemble Multi-Fidelity Neural Network for Fast Data Assimilation for Subsurface Flow in Porous Media | |
| CN108362307B (en) | Method for determining principal component factors of star sensor on-orbit attitude measurement accuracy | |
| CN113191076B (en) | Method for constructing deep learning loss function based on mineralization law | |
| Likanapaisal | Statistical moment equations for forward and inverse modeling of multiphase flow in porous media | |
| Zhu et al. | A robust inversion of logging-while-drilling responses based on deep neural network | |
| CN113608258A (en) | Self-consistent deep learning method for constructing high-resolution wave impedance inversion label | |
| Zhang | Ensemble methods of data assimilation in porous media flow for non-Gaussian prior probability density | |
| Zhao et al. | Physics-driven Deep Learning Pixel-Based Inversion of Logging-While-Drilling in Anisotropic Formation | |
| Lu | High-Performance Computing Methods for Electromagnetic Well Logging Inverse Problems | |
| CN118171582B (en) | Electromagnetic wave logging inversion method and system for azimuth while drilling by combining residual neural network and L-M algorithm |
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 |