CN118296307A - Logging data noise reduction method and device, computer equipment and storage medium - Google Patents
Logging data noise reduction method and device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN118296307A CN118296307A CN202410676009.8A CN202410676009A CN118296307A CN 118296307 A CN118296307 A CN 118296307A CN 202410676009 A CN202410676009 A CN 202410676009A CN 118296307 A CN118296307 A CN 118296307A
- Authority
- CN
- China
- Prior art keywords
- data
- logging data
- time sequence
- denoising
- characteristic
- 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
- 230000009467 reduction Effects 0.000 title claims abstract description 40
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000003860 storage Methods 0.000 title claims abstract description 7
- 238000012545 processing Methods 0.000 claims abstract description 62
- 230000007246 mechanism Effects 0.000 claims abstract description 37
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 32
- 238000000605 extraction Methods 0.000 claims abstract description 14
- 238000003786 synthesis reaction Methods 0.000 claims abstract description 10
- 238000004364 calculation method Methods 0.000 claims description 16
- 238000007781 pre-processing Methods 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 8
- 238000007499 fusion processing Methods 0.000 claims description 7
- 230000002159 abnormal effect Effects 0.000 claims description 6
- 230000006698 induction Effects 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 238000011049 filling Methods 0.000 claims description 5
- 230000000875 corresponding effect Effects 0.000 description 21
- 238000012549 training Methods 0.000 description 14
- 238000009826 distribution Methods 0.000 description 9
- 230000006870 function Effects 0.000 description 7
- 230000008569 process Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 238000005259 measurement Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 4
- 230000006835 compression Effects 0.000 description 4
- 238000007906 compression Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 4
- 238000012935 Averaging Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 3
- 238000012937 correction Methods 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 238000012163 sequencing technique Methods 0.000 description 3
- 230000001052 transient effect Effects 0.000 description 3
- 238000012512 characterization method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 239000004215 Carbon black (E152) Substances 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 239000004020 conductor Substances 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 229930195733 hydrocarbon Natural products 0.000 description 1
- 150000002430 hydrocarbons Chemical class 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 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
- 238000005457 optimization Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 239000011435 rock Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Landscapes
- Geophysics And Detection Of Objects (AREA)
Abstract
The application provides a logging data noise reduction method, a device, computer equipment and a storage medium, wherein the method comprises the following steps: acquiring time sequence logging data; inputting the time sequence logging data into a first linear layer for feature extraction to obtain first feature data; inputting the first characteristic data into a second linear layer, and reducing the dimension of the characteristic quantity dimension to obtain second characteristic data; inputting the first characteristic data into a soft threshold learning module to obtain a denoising soft threshold; the soft threshold learning module identifies redundant components based on a multi-head attention mechanism and predicts a denoising soft threshold; processing the second characteristic data according to the denoising soft threshold value to obtain third characteristic data; and summing the time sequence logging data with third characteristic data after passing through the residual connecting layer, and performing dimension reduction synthesis on the summation result to obtain the denoising stratum resistivity with the characteristic quantity of 1. The method can effectively reduce noise and redundancy of the multi-source time sequence logging data.
Description
Technical Field
The present application relates to the field of logging technologies, and in particular, to a logging data noise reduction method, a management platform, a computer device, and a storage medium.
Background
With the continuous deep exploration and development of oil and gas, the downhole logging technology plays a key role in solving the problems of oil layer and water layer definition, perforation planning and the like. However, due to engineering progress and other factors, some wells do not have open hole resistivity logging prior to completion, which results in difficulties in accurately locating the reservoir and water layers during the production phase. Under the condition of missing open hole resistivity logging, a plurality of wells have deviation in perforation, and the distribution condition of an oil layer and a water layer cannot be accurately judged. To solve this problem, conventional techniques may acquire resistivity information via transient electromagnetic logging in place of the missing true resistivity curves. However, transient electromagnetic logging is susceptible to a variety of interference factors, and the log data is composed of a series of data, with redundant portions between different data, increasing the complexity of noise processing, resulting in effective signals being masked or distorted.
Disclosure of Invention
The present application aims to solve at least one of the above technical drawbacks, and in particular, to solve the problems of high redundancy of logging data, low signal to noise ratio and ineffective noise reduction in the prior art.
In a first aspect, the present application provides a method for denoising logging data, comprising:
Acquiring time sequence logging data; the time sequence logging data comprises formation resistivity and a plurality of other logging data, and the dimension of the time sequence logging data comprises a time sequence length and a feature quantity;
inputting the time sequence logging data into a first linear layer for feature extraction to obtain first feature data;
Inputting the first characteristic data into a second linear layer, and reducing the dimension of the characteristic quantity dimension to obtain second characteristic data;
inputting the first characteristic data into a soft threshold learning module to obtain a denoising soft threshold; the soft threshold learning module identifies redundant components based on a multi-head attention mechanism and predicts a denoising soft threshold;
processing the second characteristic data according to the denoising soft threshold value to obtain third characteristic data;
and summing the time sequence logging data with third characteristic data after passing through the residual connecting layer, and performing dimension reduction synthesis on the summation result to obtain the denoising stratum resistivity with the characteristic quantity of 1.
In one embodiment, inputting the first feature data into a soft threshold learning module to obtain a denoising soft threshold value includes:
absolute value and global average value of the first characteristic data are pooled to obtain non-time sequence characteristic data;
Inputting the non-time sequence characteristic data into a third linear layer, and reducing the characteristic quantity dimension to 1 to obtain fourth characteristic data;
inputting the non-time sequence characteristic data into a multi-head attention mechanism unit to obtain multi-head attention splicing weight;
And sequentially processing the multi-head attention splicing weight through a batch normalization layer, a ReLu layer, a second full-connection layer and a Sigmoid layer to obtain fifth characteristic data, and multiplying the fifth characteristic data with fourth characteristic data to obtain a denoising soft threshold.
In one embodiment, inputting non-time sequence characteristic data into a multi-head attention mechanism unit to obtain multi-head attention splicing weight, including:
Converting the non-time sequence characteristic data into corresponding calculation parameters through three linear layers respectively;
Inputting each calculation parameter into a linear layer corresponding to each attention head respectively to obtain multi-head attention characteristics corresponding to each attention head;
And (3) after the multi-head attention features are subjected to scaling dot product processing and fusion processing, obtaining multi-head attention splicing weights.
In one embodiment, the other well logging data includes sonic logging data, natural gamma logging data, induction logging data, natural potential logging data, and through-casing electromagnetic logging data from a plurality of different ammeter tests.
In one embodiment, the residual connection layer is a fourth linear layer.
In one embodiment, before the time sequence logging data is input into the first linear layer to perform feature extraction, the method further includes:
Preprocessing the time sequence logging data.
In one embodiment, the pre-processing of the sequencing log data includes any one or more of the following:
correcting abnormal values in the time sequence logging data;
Filling in missing values in the time series logging data.
In a second aspect, the present application provides a logging data denoising apparatus, comprising:
The data acquisition module is used for acquiring time sequence logging data; the time sequence logging data comprises formation resistivity and a plurality of other logging data, and the dimension of the time sequence logging data comprises a time sequence length and a feature quantity;
The first processing module is used for inputting the time sequence logging data into the first linear layer to perform feature extraction so as to obtain first feature data;
The second processing module is used for inputting the first characteristic data into the second linear layer, and reducing the dimension of the characteristic quantity dimension to obtain second characteristic data;
the third processing module is used for inputting the first characteristic data into the soft threshold learning module to obtain a denoising soft threshold; the soft threshold learning module identifies redundant components based on a multi-head attention mechanism and predicts a denoising soft threshold;
The fourth processing module is used for processing the second characteristic data according to the denoising soft threshold value to obtain third characteristic data;
And the fifth processing module is used for summing the time sequence logging data with the third characteristic data after passing through the residual connection layer, and performing dimension reduction synthesis on the summation result to obtain the denoising stratum resistivity with the characteristic quantity of 1.
In a third aspect, the present application provides a computer device comprising one or more processors and a memory having stored therein computer readable instructions which, when executed by the one or more processors, perform the steps of the logging data denoising method of any of the embodiments described above.
In a fourth aspect, the present application provides a storage medium having stored therein computer readable instructions which, when executed by one or more processors, cause the one or more processors to perform the steps of the logging data denoising method of any of the embodiments described above.
From the above technical solutions, the embodiment of the present application has the following advantages:
Firstly, time sequence logging data containing various logging data are obtained as input, then the input data are subjected to primary feature conversion through a first linear layer, and then feature quantity dimension is subjected to dimension reduction through a second linear layer, so that feature compression and redundancy reduction are realized. The output of the first linear layer is then input into a soft threshold learning module, which captures the relevant characteristics of the input data using a multi-headed attention mechanism and predicts a de-noised soft threshold. And applying the predicted soft threshold value to the second characteristic data to complete the actual soft threshold value denoising and redundancy removing processing. And finally, combining the original input information with the denoised characteristic data through residual connection, and further reducing the dimension to synthesize single-channel output, thus obtaining the final denoised stratum resistivity. The scheme combines a multi-head attention mechanism, a soft thresholding attention mechanism and a residual error network, so that the training effectiveness can be ensured, the dimension reduction compression in the scheme also effectively reduces the data redundancy and noise interference, the feature distribution is more compact and efficient, and the calculation complexity is reduced. And the correlation and the importance among the features can be better captured, so that the characterization capability of the features is improved. Meanwhile, the multi-head attention mechanism can better capture the redundant part and the time sequence noise random fluctuation part in the process of learning the soft threshold value, so that the corresponding denoising soft threshold value is predicted, and finally, the soft threshold value processing unit realizes denoising, thereby having strong redundancy removal and noise reduction capability. And finally, single-channel results are produced, the method focuses on predicting a precise denoising stratum resistivity curve, and the core requirements of logging exploration are met.
Drawings
In order to more clearly illustrate the embodiments of the application or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are only some embodiments of the application, and that other drawings can be obtained from these drawings without inventive faculty for a person skilled in the art.
FIG. 1 is a flow chart of a method for denoising logging data according to an embodiment of the present application;
FIG. 2 is a schematic diagram of a structure for logging data noise reduction according to an embodiment of the present application;
FIG. 3 is a flow chart of predicting a soft threshold for denoising according to one embodiment of the present application;
FIG. 4 is a schematic diagram of a structure for noise reduction of log data according to another embodiment of the present application;
FIG. 5 is a block diagram of a logging data noise reduction apparatus according to an embodiment of the present application;
Fig. 6 is an internal structural diagram of a computer device according to an embodiment of the present application.
Detailed Description
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
In the field of logging, different types of logging tools can acquire different measurement data. These time series acquired measurement data are combined to form a multi-dimensional time series logging data set. These data are susceptible to electromagnetic interference, electrical noise, etc. of the measuring device itself. Substances in the subsurface medium, such as rock, water, etc., can also interfere with and attenuate signals. More importantly, the test is also affected by the sleeve, which is made of conductive material, which introduces additional interference signals into the various response signals and distortion. The correction of the casing effect needs to involve complex models and algorithms, the diversity of actual geological conditions also leads to increased correction difficulty, and many correction algorithms only stay at the theoretical level and cannot be applied to actual working conditions. Therefore, various types of over-casing transient electromagnetic logging curves belong to multi-source data with low signal-to-noise ratio and high redundancy. The effect of noise is more pronounced in data with low signal to noise ratio. Highly redundant data also adds complexity to noise processing because similar noise patterns may exist in the redundant data, making noise removal more challenging. Therefore, in order to better utilize the logging data, the present embodiment provides a logging data denoising method based on deep learning and neural network, which includes steps S102 to S112.
S102, acquiring time sequence logging data. The time series logging data includes formation resistivity and a plurality of other logging data, and the dimensions of the time series logging data include a time series length and a feature quantity.
It is understood that time series logging data refers to a series of time-varying response data obtained by a variety of different logging means during a logging process. The time series logging data includes formation resistivity and a plurality of other logging data. Formation resistivity, which refers to the degree of resistance of the formation to the passage of current, is a key parameter describing the conductivity of the formation. Accurate formation resistivity acquisition is of great importance in identifying hydrocarbon reservoirs. Other logging data are data obtained by other logging means in addition to formation resistivity, such as curve data obtained by various measurement means such as sonic logging, density logging, natural gamma logging, electrical logging, etc. Regarding the dimension of the time series logging data, the feature quantity dimension represents the number of kinds of measurement data in the time series logging data. For example, the time series logging data is 40 data from the group consisting of AC (acoustic logging), GR (natural gamma curve), CON (induction logging), SP (natural potential curve), through-casing electromagnetic logging STEMUB-STEMUB (readings of instrument ammeter 01-35), RT (formation resistivity), and the feature quantity is 40. The time series length represents the number of different time points included in each measurement data. In order to improve the processing efficiency, all data is not processed at one time in the field of the neural network, but is processed in batches. Thus, training batch dimensions may also be added to these two dimensions. As shown in fig. 2, the data in this embodiment may include B, S, F dimensions, where B represents the training batch, S represents the time series length, and F represents the feature quantity. The following will take the example of having a training batch dimension, but this dimension may not be included in actual use.
S104, inputting the time sequence logging data into the first linear layer for feature extraction to obtain first feature data.
It will be appreciated that the method of the present embodiment is applicable to the neural network architecture of fig. 2. The first linear layer is used for carrying out linear transformation on input and output through a weight matrix to realize preliminary coding and feature conversion, and extracting new feature representations which are more relevant to stratum conductivity from original time sequence logging data, so that preparation is made for subsequent feature extraction and processing of deeper layers. The first linear layer has no influence on the dimension of the time sequence logging data, the first characteristic data is the same as the dimension of the time sequence logging data, as shown in fig. 2, the dimension of the data input into the first linear layer is (B, S, F), and the dimension of the data output by the first linear layer is (B, S, F).
S106, inputting the first characteristic data into a second linear layer, and reducing the dimension of the characteristic quantity dimension to obtain second characteristic data.
It can be understood that the second linear layer is used for reducing the dimension of the feature data of the first layer in the dimension of the feature quantity, and redundant information can be reduced through dimension reduction, so that feature distribution is more compact. That is, the second linear layer only affects the first feature data in a feature quantity dimension, the second feature data being the same as the first feature data in a training batch, time series length dimension, and the feature quantity dimension being smaller than the first feature data. As shown in fig. 2, the first characteristic data outputted by the first linear layer has dimensions (B, S, F), the second characteristic data outputted after passing through the second linear layer has dimensions (B, S, H), and H is a positive integer smaller than F. Excessive dimensionality reduction of the second linear layer may result in loss of effective information, so the ratio between H and F needs to be weighed by experiment. May be generally set to 1/4 to 1/3. For example, the first feature data may have dimensions (32, 40, 42) and the second feature data may have dimensions (32, 40, 12).
S108, inputting the first characteristic data into a soft threshold learning module to obtain a denoising soft threshold.
It will be appreciated that the architecture of fig. 2 is an improvement over conventional depth residual shrink networks, which are used for the image domain, which the present embodiment improves to be suitable for use in time-series data processing. And further combines the soft threshold mechanism therein with the soft threshold mechanism. The soft threshold learning module in this embodiment is configured to identify redundant components in input data based on a multi-headed attention mechanism and predict a denoising soft threshold. The multi-headed attention mechanism focuses on the higher-attention-scoring portion of the input data, which is the redundant portion in the multi-dimensional feature, based on which a denoising soft threshold can be predicted, and subsequently the redundant portion can be removed by the soft threshold mechanism. In addition, in the soft threshold learning module, the time sequence dimensions can be fused, random noise fluctuation of the time sequence dimensions can be found through comparison, and related information about the random noise fluctuation is learned together when the soft threshold is predicted, so that noise can be removed together when the subsequent soft threshold is processed. For different time sequence logging data, the soft threshold learning module can drive and learn the optimal threshold independently according to the data distribution of the data input currently, manual setting is not needed, and the multi-head attention mechanism can well capture the correlation and importance among features, so that the soft threshold with good noise reduction and redundancy elimination effects is accurately predicted. The output of the soft threshold learning module is used for each group of data in the batch data, so that the size of the characteristic number dimension and the time sequence length dimension are both 1, as shown in fig. 2, the first characteristic data dimension input into the soft threshold learning module is (B, S, F), and the dimension of the denoising soft threshold output by the soft threshold learning module is (B, 1). Specifically, the first feature data may have a dimension of (32, 40, 42), and the denoising soft threshold may have a dimension of (32,1,1).
S110, processing the second characteristic data according to the denoising soft threshold value to obtain third characteristic data.
It is understood that a soft threshold mechanism refers to filtering an input signal with a flexibly variable threshold. The soft threshold learning module in this embodiment learns a soft threshold outputted after distribution of input data, that is, a soft threshold that varies according to the input data. The soft threshold processing is to set the feature with absolute value lower than the denoising soft threshold in the input data to zero, and other features are adjusted to zero accordingly to realize shrinkage. The expression of the soft threshold function and its solution are as follows:
；
；
X in the above formula represents the second characteristic data, τ represents the denoising soft threshold value, and y represents the denoised third characteristic data. The step applies the denoising soft threshold value obtained in the previous step to the second characteristic data to finish actual denoising, and redundant characteristics and noise interference are effectively removed. The step does not affect the dimension of the data, only the specific value of the data. Therefore, the dimensions of the second feature data and the third feature data are the same, and as shown in fig. 2, the dimensions of the second feature data outputted by the second linear layer are (B, S, H), and the dimensions of the third feature data obtained after the soft threshold are (B, S, H), for example, the dimensions of the second feature data are (32, 40, 12), and the dimensions of the third feature data may be (32, 40, 12).
And S112, summing the time sequence logging data with third characteristic data after passing through the residual connection layer, and performing dimension reduction synthesis on the summation result to obtain the denoising stratum resistivity with the characteristic quantity of 1.
It can be understood that the sequential characteristic data sequentially passes through the first linear layer, the second linear layer and the soft threshold processing unit to be a main branch of data processing, but as the neural network deepens, problems such as gradient disappearance and gradient explosion are brought simultaneously, and a lot of troubles are brought to model training. To solve this problem, the original input, i.e. the sequencing log data, is summed with the output of the main leg by the residual connection layer, a residual network structure may be formed. The structure can well solve the problem of gradient dissipation, can also enhance the expression effect of the model and reduce the training difficulty of the model. The residual connection layer adjusts the dimension of the time series logging data to be the same as the third characteristic data so as to facilitate better summation with the third characteristic data. As shown in fig. 2, the dimension of the time series logging data is (B, S, F), and the dimension of the third feature data obtained after the soft threshold is (B, S, H), the residual connection layer reduces the dimension of the time series logging data to (B, S, H).
For example, the time series logging data has a dimension (32, 40, 42), the third feature data has a dimension (32, 40, 12), and the data output by the residual connection layer has a dimension (32, 40, 12). Finally, the denoised multichannel data is required to be fused, and then the feature quantity dimension is reduced to 1, namely only the denoised formation resistivity of a single channel is output, only a single channel result is produced, and the focusing model is a main task of predicting the formation resistivity. The dimension reduction synthesis here is typically implemented using a fully connected layer. The function of the full connection layer is to sum the multi-channel data of the feature quantity dimension according to the weight obtained by training, and the multi-channel data becomes final single-channel data. And finally, the obtained denoising stratum resistivity is still time sequence data. The dimension of the training batch is the same as the dimension of the length of the time sequence between the summation result and the result output by the full connection layer, and only the feature quantity dimension is fused. Specifically, if the dimension of the summation result is (B, S, H), the dimension of the result becomes (B, S, 1) after passing through the full connection layer. For example, the dimension of the summation result is (32, 40, 12), and the dimension of the final denoising resistivity is (32, 40, 1).
Firstly, time sequence logging data containing various logging data are obtained as input, then the input data are subjected to primary feature conversion through a first linear layer, and then feature quantity dimension is subjected to dimension reduction through a second linear layer, so that feature compression and redundancy reduction are realized. The output of the first linear layer is then input into a soft threshold learning module, which captures the relevant characteristics of the input data using a multi-headed attention mechanism and predicts a de-noised soft threshold. And applying the predicted soft threshold value to the second characteristic data to complete the actual soft threshold value denoising and redundancy removing processing. And finally, combining the original input information with the denoised characteristic data through residual connection, and further reducing the dimension to synthesize single-channel output, thus obtaining the final denoised stratum resistivity. The scheme combines a multi-head attention mechanism, a soft thresholding attention mechanism and a residual error network, so that the training effectiveness can be ensured, the dimension reduction compression in the scheme also effectively reduces the data redundancy and noise interference, the feature distribution is more compact and efficient, and the calculation complexity is reduced. And the correlation and the importance among the features can be better captured, so that the characterization capability of the features is improved. Meanwhile, the multi-head attention mechanism can better capture the redundant part and the time sequence noise random fluctuation part in the process of learning the soft threshold value, so that the corresponding denoising soft threshold value is predicted, and finally, the soft threshold value processing unit realizes denoising, thereby having strong redundancy removal and noise reduction capability. And finally, single-channel results are produced, the method focuses on predicting a precise denoising stratum resistivity curve, and the core requirements of logging exploration are met.
The above-described embodiments need to be implemented in dependence on the log data noise reduction model in fig. 2, which includes a first linear layer, a second linear layer, a soft threshold learning module, a soft threshold processing unit, a residual connection layer, a summing unit, and a full connection layer. The first linear layer is used for carrying out preliminary feature extraction on time sequence logging data, so that preliminary coding and feature conversion are realized, and new feature representation which is more relevant to stratum conductivity is extracted from the original time sequence logging data. The first linear layer is respectively connected with the second linear layer and the soft threshold learning module. The second linear layer is used for reducing the dimension of the first characteristic data, redundant information can be reduced through dimension reduction, and the characteristic distribution is more compact. The soft threshold learning module is used for identifying redundant components in input data based on a multi-head attention mechanism and predicting a denoising soft threshold. The soft threshold learning module and the second linear layer are both connected with the soft threshold processing unit. And the soft threshold processing unit carries out soft thresholding on the second characteristic data output by the second linear layer according to the denoising soft threshold output by the soft threshold learning module, so as to complete the actual denoising and redundancy removing processes. The soft threshold processing unit is connected with the summing unit, and the input of the whole model is also connected with the summing unit through a residual error connecting layer to form a residual error network structure. And the third characteristic data after soft thresholding and the time sequence logging data passing through the residual connecting layer are input into a summation unit together for summation. The summing unit is connected with the full connection layer. And the full-connection layer carries out fusion dimension reduction on the summation result in the characteristic number dimension, and finally fuses the multi-channel data in the characteristic number dimension into the single-channel denoising stratum resistivity.
The training of the model is similar to the training mode of a traditional depth residual error shrinkage network and a multi-head attention mechanism. And collecting a plurality of different time sequence logging data as a training set, and taking the corresponding standard formation resistivity as a label. Since the objective is to predict accurate denoising formation resistivity, regression loss functions, commonly used Mean Square Error (MSE), smooth L1 loss and the like can be used, the denoising resistivity predicted by the model is compared with the marked standard formation resistivity, and a loss value is calculated. After initializing all weights in the model, using a small batch random gradient descent optimization algorithm, performing forward calculation to obtain denoising resistivity output by the model, calculating loss of a labeling value, and updating parameters of each layer according to back propagation of the loss value. Strategies such as learning rate decay, momentum, etc. can also be used to accelerate convergence. The soft threshold learning module is a sub-module that requires end-to-end training with the entire network. Finally, the model with the performance index meeting the requirements is obtained by adjusting the weight of the model, the parameters of the optimizer, the number of layers and the like.
In one embodiment, referring to fig. 3, the first feature data is input to the soft threshold learning module to obtain the denoising soft threshold, which includes steps S302 to S308. In addition, referring to fig. 4, the soft threshold learning module includes a pool layer for absolute value and global average, a multi-head attention mechanism unit, a batch normalization layer, reLu layers, a second full-connection layer, and a Sigmoid layer.
S302, absolute value and global average value of the first characteristic data are pooled to obtain non-time sequence characteristic data.
It will be appreciated that the absolute value takes the absolute value of the data, i.e. the sign of the value is cancelled. And carrying out average pooling operation on the whole feature map by global average pooling, and compressing a certain dimension of the whole feature into a single-channel scalar value. In this embodiment, global averaging is to reduce the dimension of the multi-channel data in the time sequence length dimension of the first feature data into single-channel data, so that the single-channel data is changed from the original time sequence data into non-time sequence data, and the global features of the first feature data are extracted, so as to provide a feature basis for the comparison of the multi-dimensional redundant components by the subsequent multi-head attention mechanism unit. Assuming that the first feature data dimension is (B, S, F), the non-temporal feature data output by the absolute value and global averaging layer has dimensions of (B, 1, F). Specifically, the dimension of the first feature data may be (32, 40, 42), and the dimension of the denoising soft threshold may be (32, 1, 42).
S304, inputting the non-time sequence characteristic data into a third linear layer, and reducing the dimension of the characteristic quantity to 1 to obtain fourth characteristic data.
It will be appreciated that the third linear layer further compresses the non-time series characteristic data so that it has only one unit value in both the number of characteristics and the time series length. This amounts to focusing the entire feature on a scalar. Through the extreme dimension reduction operation, the model can be forced to learn the statistical distribution characteristics of the whole input data. Assuming that the dimensions of the non-time series feature data output by the absolute value and global averaging layers are (B, 1, f), the dimensions of the fourth feature data output by the third linear layer are (B, 1). Specifically, the first feature data may have a dimension of (32, 1, 42), and the denoising soft threshold may have a dimension of (32,1,1).
S306, inputting the non-time sequence characteristic data into a multi-head attention mechanism unit to obtain multi-head attention splicing weight.
It will be appreciated that the multi-headed attention mechanism unit may divide the attention into a plurality of subspaces, with attention calculations being performed within each subspace, so that correlations between inputs may be captured from different representation subspaces. For each subspace, the attention weight reflects the degree to which the current location feature is correlated with all other location features. By stitching these subspace weights, a attention representation can be derived that fully describes the overall input distribution. Specifically, the multi-head attention mechanism unit includes a plurality of linear layers, and the non-time sequence feature data can be converted into three calculation parameters of Query, key and Value through the linear layers corresponding to the Query, key and Value respectively. The three calculation parameters are respectively input into the corresponding linear layers in the plurality of attention heads, and the weight in the corresponding linear layer of each attention head is different, so that a plurality of groups of multi-head attention features corresponding to the attention heads and different in output are obtained. After the multi-head attention features of each group are subjected to scaling dot product processing and fusion processing (concat), multi-head attention splicing weights can be obtained. The feature number dimension of the multi-head attention splicing weight obtained after fusion processing is three times of the input, the dimension of non-time sequence feature data is (B, 1, F), and the dimension of the multi-head attention splicing weight is (B, 1,3F). Specifically, the dimension of the first feature data may be (32, 1, 42), and the dimension of the denoising soft threshold may be (32, 1, 126).
And S308, sequentially processing the multi-head attention splicing weight through a batch normalization layer, a ReLu layer, a second full-connection layer and a Sigmoid layer to obtain fifth characteristic data, and multiplying the fifth characteristic data by fourth characteristic data to obtain a denoising soft threshold.
It can be understood that the batch normalization layer is used for normalizing batch data, so that the convergence rate of a model can be increased, and the gradient dispersion problem in a deep network can be relieved to a certain extent. The ReLu layer is a layer that is non-linearly mapped using a ReLu activation function, the ReLu activation function has the expression max (0,I), and I represents the input of the ReLu layer. The ReLu layers may better produce feature selection. The second full connection layer synthesizes the features after ReLu layers are selected and activated, and combines the multi-channel features in the feature quantity dimension into one channel. Finally, through a Sigmoid activation function layer, mapping of f (x) =1/(1+e-x) is realized, and fifth characteristic data is obtained. The fifth characteristic data and the fourth characteristic data are multiplied to obtain a denoising soft threshold value. The soft threshold for denoising is not only a positive number, but also not too large, i.e., the soft thresholded features are not all zero.
In one embodiment, the residual connection layer is a fourth linear layer. Because the dimension of the third characteristic data is different from the dimension of the logging time sequence characteristic data after the dimension reduction processing, identity mapping cannot be directly used. In order to construct the residual network, the fourth linear layer is used to reduce the dimension of the sequenced logging data, so that the dimension of the sequenced logging data is kept the same as the dimension of the third characteristic data, and the sequenced logging data can be directly summed in the summing unit.
In one embodiment, before the time sequence logging data is input into the first linear layer to perform feature extraction, the method further includes: preprocessing the time sequence logging data.
It can be appreciated that the time sequence logging data may have the conditions of non-uniform format, missing data, abnormal data and the like which do not conform to the model input conditions in the actual logging process. Therefore, the time sequence logging data can be preprocessed before being input into the first linear layer, so that the time sequence logging data meets the input requirement. Specifically, the sequencing log data is pre-processed, including any one or more of: outliers in the time series log data are corrected. Filling in missing values in the time series logging data.
In one embodiment, because the time series logging data is logging data continuously acquired at certain time intervals, there is a strong correlation between the data at adjacent times. Thus, filling in missing values may utilize interpolation methods, such as linear interpolation, spline interpolation, etc., to estimate missing values from values of neighboring data points. In addition, when interpolation is performed, weighting the values of adjacent data points can be considered, and adjacent data points which are closer to the missing data are given greater weight. Statistical properties of the log, such as mean, median, etc., may also be used, with the missing values replaced with statistical values. In addition, a regression model can be established based on the values of other related logging curves to estimate the missing values. Finally, if the amount of data is sufficient, machine learning methods, such as K-nearest neighbors, decision trees, etc., can also be used to predict the missing values based on the known data.
In one particular embodiment, the outlier removal may be based on empirical rules, setting a threshold interval, and values outside the interval are considered outliers. The outliers exceeding 3 standard deviations from the mean can also be considered outliers using the three sigma principle. Further, a value that is too far from the upper and lower quartiles may be regarded as an outlier or the like based on the box plot analysis.
The application provides a logging data noise reduction device, referring to fig. 5, comprising a data acquisition module 510, a first processing module 520, a second processing module 530, a third processing module 540, a fourth processing module 550 and a fifth processing module 560.
The data acquisition module 510 is configured to acquire time-series logging data. The time series logging data includes formation resistivity and a plurality of other logging data, and the dimensions of the time series logging data include a time series length and a feature quantity.
The first processing module 520 is configured to input the time-series logging data into the first linear layer for feature extraction, so as to obtain first feature data.
The second processing module 530 is configured to input the first feature data into the second linear layer, and dimension-reduce the feature quantity dimension to obtain second feature data.
The third processing module 540 is configured to input the first feature data into the soft threshold learning module to obtain a denoising soft threshold. The soft threshold learning module identifies redundant components based on a multi-headed attention mechanism and predicts a denoising soft threshold.
The fourth processing module 550 is configured to process the second feature data according to the soft threshold of the denoising soft threshold to obtain third feature data.
The fifth processing module 560 is configured to sum the time-series logging data with the third feature data after passing through the residual connection layer, and perform dimension reduction synthesis on the sum result to obtain the denoised formation resistivity with the feature quantity of 1.
In one embodiment, the third processing module 540 is configured to perform absolute value and global mean pooling on the first feature data to obtain non-time-series feature data; inputting the non-time sequence characteristic data into a third linear layer, and reducing the characteristic quantity dimension to 1 to obtain fourth characteristic data; inputting the non-time sequence characteristic data into a multi-head attention mechanism unit to obtain multi-head attention splicing weight; and sequentially processing the multi-head attention splicing weight through a batch normalization layer, a ReLu layer, a second full-connection layer and a Sigmoid layer to obtain fifth characteristic data, and multiplying the fifth characteristic data with fourth characteristic data to obtain a denoising soft threshold.
In one embodiment, the third processing module 540 is configured to convert the non-time-series characteristic data into corresponding calculation parameters through three linear layers respectively; inputting each calculation parameter into a linear layer corresponding to each attention head respectively to obtain multi-head attention characteristics corresponding to each attention head; and (3) after the multi-head attention features are subjected to scaling dot product processing and fusion processing, obtaining multi-head attention splicing weights.
In one embodiment, the other well logging data includes sonic logging data, natural gamma logging data, induction logging data, natural potential logging data, and through-casing electromagnetic logging data from a plurality of different ammeter tests.
In one embodiment, the residual connection layer is a fourth linear layer.
In one embodiment, the logging data denoising module further comprises a preprocessing module. The preprocessing module is used for preprocessing the time sequence logging data.
In one embodiment, the preprocessing module is configured to perform preprocessing by any one or more of: correcting abnormal values in the time sequence logging data; filling in missing values in the time series logging data.
For specific limitations of the logging data denoising module, reference may be made to the above limitation of the logging data denoising method, and no further description is given here. The modules in the logging data noise reduction device can be all or partially implemented by software, hardware and a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules. It should be noted that, in the embodiment of the present application, the division of the modules is schematic, which is merely a logic function division, and other division manners may be implemented in actual implementation.
The present application provides a computer device comprising one or more processors, and a memory having stored therein computer readable instructions that, when executed by the one or more processors, perform: acquiring time sequence logging data; the time sequence logging data comprises formation resistivity and a plurality of other logging data, and the dimension of the time sequence logging data comprises a time sequence length and a feature quantity; inputting the time sequence logging data into a first linear layer for feature extraction to obtain first feature data; inputting the first characteristic data into a second linear layer, and reducing the dimension of the characteristic quantity dimension to obtain second characteristic data; inputting the first characteristic data into a soft threshold learning module to obtain a denoising soft threshold; the soft threshold learning module identifies redundant components based on a multi-head attention mechanism and predicts a denoising soft threshold; processing the second characteristic data according to the denoising soft threshold value to obtain third characteristic data; and summing the time sequence logging data with third characteristic data after passing through the residual connecting layer, and performing dimension reduction synthesis on the summation result to obtain the denoising stratum resistivity with the characteristic quantity of 1.
In one embodiment, computer-readable instructions, when executed by one or more processors, perform: absolute value and global average value of the first characteristic data are pooled to obtain non-time sequence characteristic data; inputting the non-time sequence characteristic data into a third linear layer, and reducing the characteristic quantity dimension to 1 to obtain fourth characteristic data; inputting the non-time sequence characteristic data into a multi-head attention mechanism unit to obtain multi-head attention splicing weight; and sequentially processing the multi-head attention splicing weight through a batch normalization layer, a ReLu layer, a second full-connection layer and a Sigmoid layer to obtain fifth characteristic data, and multiplying the fifth characteristic data with fourth characteristic data to obtain a denoising soft threshold.
In one embodiment, computer-readable instructions, when executed by one or more processors, perform: converting the non-time sequence characteristic data into corresponding calculation parameters through three linear layers respectively; inputting each calculation parameter into a linear layer corresponding to each attention head respectively to obtain multi-head attention characteristics corresponding to each attention head; and (3) after the multi-head attention features are subjected to scaling dot product processing and fusion processing, obtaining multi-head attention splicing weights.
In one embodiment, the other well logging data includes sonic logging data, natural gamma logging data, induction logging data, natural potential logging data, and through-casing electromagnetic logging data from a plurality of different ammeter tests.
In one embodiment, the residual connection layer is a fourth linear layer.
In one embodiment, computer-readable instructions, when executed by one or more processors, perform: preprocessing the time sequence logging data.
In one embodiment, computer-readable instructions, when executed by one or more processors, perform: correcting abnormal values in the time sequence logging data; and/or fill in missing values in the time series log data.
Schematically, as shown in fig. 6, fig. 6 is a schematic internal structure of a computer device according to an embodiment of the present application. Referring to FIG. 6, a computer device 600 includes a processing component 602 that further includes one or more processors and memory resources represented by a memory 601 for storing instructions, such as applications, executable by the processing component 602. The application programs stored in the memory 601 may include one or more modules, each corresponding to a set of instructions. Further, the processing component 602 is configured to execute instructions to perform the steps of the logging data denoising method of any of the embodiments described above.
The computer device 600 may also include a power component 603 configured to perform power management of the computer device 600, a wired or wireless network interface 604 configured to connect the computer device 600 to a network, and an input output (I/O) interface 605.
It will be appreciated by those skilled in the art that the structure shown in FIG. 6 is merely a block diagram of some of the structures associated with the present inventive arrangements and is not limiting of the computer device to which the present inventive arrangements may be applied, and that a particular computer device may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.
The present application provides a storage medium having stored therein computer readable instructions that, when executed by one or more processors, cause the one or more processors to perform: acquiring time sequence logging data; the time sequence logging data comprises formation resistivity and a plurality of other logging data, and the dimension of the time sequence logging data comprises a time sequence length and a feature quantity; inputting the time sequence logging data into a first linear layer for feature extraction to obtain first feature data; inputting the first characteristic data into a second linear layer, and reducing the dimension of the characteristic quantity dimension to obtain second characteristic data; inputting the first characteristic data into a soft threshold learning module to obtain a denoising soft threshold; the soft threshold learning module identifies redundant components based on a multi-head attention mechanism and predicts a denoising soft threshold; processing the second characteristic data according to the denoising soft threshold value to obtain third characteristic data; and summing the time sequence logging data with third characteristic data after passing through the residual connecting layer, and performing dimension reduction synthesis on the summation result to obtain the denoising stratum resistivity with the characteristic quantity of 1.
In one embodiment, computer-readable instructions, when executed by one or more processors, cause the one or more processors to perform: absolute value and global average value of the first characteristic data are pooled to obtain non-time sequence characteristic data; inputting the non-time sequence characteristic data into a third linear layer, and reducing the characteristic quantity dimension to 1 to obtain fourth characteristic data; inputting the non-time sequence characteristic data into a multi-head attention mechanism unit to obtain multi-head attention splicing weight; and sequentially processing the multi-head attention splicing weight through a batch normalization layer, a ReLu layer, a second full-connection layer and a Sigmoid layer to obtain fifth characteristic data, and multiplying the fifth characteristic data with fourth characteristic data to obtain a denoising soft threshold.
In one embodiment, computer-readable instructions, when executed by one or more processors, cause the one or more processors to perform: converting the non-time sequence characteristic data into corresponding calculation parameters through three linear layers respectively; inputting each calculation parameter into a linear layer corresponding to each attention head respectively to obtain multi-head attention characteristics corresponding to each attention head; and (3) after the multi-head attention features are subjected to scaling dot product processing and fusion processing, obtaining multi-head attention splicing weights.
In one embodiment, the other well logging data includes sonic logging data, natural gamma logging data, induction logging data, natural potential logging data, and through-casing electromagnetic logging data from a plurality of different ammeter tests.
In one embodiment, the residual connection layer is a fourth linear layer.
In one embodiment, computer-readable instructions, when executed by one or more processors, cause the one or more processors to perform: preprocessing the time sequence logging data.
In one embodiment, computer-readable instructions, when executed by one or more processors, cause the one or more processors to perform: correcting abnormal values in the time sequence logging data; and/or fill in missing values in the time series log data.
In the present specification, each embodiment is described in a progressive manner, and each embodiment focuses on the difference from other embodiments, and may be combined according to needs, and the same similar parts may be referred to each other.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
1. A method of logging data noise reduction, comprising:
Acquiring time sequence logging data; the time sequence logging data comprises formation resistivity and a plurality of other logging data, and the dimension of the time sequence logging data comprises a time sequence length and a characteristic quantity;
Inputting the time sequence logging data into a first linear layer for feature extraction to obtain first feature data;
Inputting the first characteristic data into a second linear layer, and reducing the dimension of the characteristic quantity dimension to obtain second characteristic data;
inputting the first characteristic data into a soft threshold learning module to obtain a denoising soft threshold; the soft threshold learning module identifies redundant components based on a multi-head attention mechanism and predicts the denoising soft threshold;
Processing the second characteristic data according to the denoising soft threshold value to obtain third characteristic data;
and summing the time sequence logging data with the third characteristic data after passing through a residual connection layer, and performing dimension reduction synthesis on a summation result to obtain the denoising stratum resistivity with the characteristic quantity of 1.
2. The method of denoising logging data according to claim 1, wherein inputting the first feature data into a soft threshold learning module to obtain a denoising soft threshold comprises:
Absolute value and global mean value of the first characteristic data are pooled to obtain non-time sequence characteristic data;
Inputting the non-time sequence characteristic data into a third linear layer, and reducing the characteristic quantity dimension to 1 to obtain fourth characteristic data;
inputting the non-time sequence characteristic data into a multi-head attention mechanism unit to obtain multi-head attention splicing weight;
and processing the multi-head attention splicing weight through a batch normalization layer, a ReLu layer, a second full-connection layer and a Sigmoid layer in sequence to obtain fifth characteristic data, and multiplying the fifth characteristic data with the fourth characteristic data to obtain the denoising soft threshold.
3. The method of denoising logging data according to claim 2, wherein inputting the non-time series characteristic data into a multi-head attention mechanism unit to obtain multi-head attention stitching weights comprises:
converting the non-time sequence characteristic data into corresponding calculation parameters through three linear layers respectively;
Inputting each calculation parameter into a linear layer corresponding to each attention head respectively to obtain multi-head attention characteristics corresponding to each attention head;
And performing scaling dot product processing and fusion processing on each multi-head attention characteristic to obtain the multi-head attention splicing weight.
4. The method of logging data noise reduction of claim 1, wherein the other logging data comprises sonic logging data, natural gamma logging data, induction logging data, natural potential logging data, and over-casing electromagnetic logging data from a plurality of different ammeter tests.
5. The method of logging data noise reduction of claim 1, wherein the residual connection layer is a fourth linear layer.
6. The method of denoising log data according to claim 1, further comprising, before the inputting the time-series log data into the first linear layer for feature extraction, obtaining first feature data:
And preprocessing the time sequence logging data.
7. The method of logging data denoising according to claim 6, wherein the preprocessing of the time series logging data comprises any one or more of:
correcting abnormal values in the time sequence logging data;
filling up the missing value in the time sequence logging data.
8.A logging data noise reduction apparatus, comprising:
the data acquisition module is used for acquiring time sequence logging data; the time sequence logging data comprises formation resistivity and a plurality of other logging data, and the dimension of the time sequence logging data comprises a time sequence length and a characteristic quantity;
the first processing module is used for inputting the time sequence logging data into a first linear layer to perform feature extraction to obtain first feature data;
the second processing module is used for inputting the first characteristic data into a second linear layer, and reducing the dimension of the characteristic quantity to obtain second characteristic data;
The third processing module is used for inputting the first characteristic data into the soft threshold learning module to obtain a denoising soft threshold; the soft threshold learning module identifies redundant components based on a multi-head attention mechanism and predicts the denoising soft threshold;
the fourth processing module is used for processing the second characteristic data according to the denoising soft threshold value to obtain third characteristic data;
And a fifth processing module, configured to sum the time-series logging data with the third feature data after passing through the residual connection layer, and perform dimension reduction synthesis on the sum result to obtain the denoised formation resistivity with the feature quantity of 1.
9. A computer device comprising one or more processors and a memory having stored therein computer readable instructions which, when executed by the one or more processors, perform the steps of the logging data denoising method of any one of claims 1 to 7.
10. A storage medium having stored therein computer readable instructions which, when executed by one or more processors, cause the one or more processors to perform the steps of the logging data denoising method of any of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410676009.8A CN118296307A (en) | 2024-05-29 | 2024-05-29 | Logging data noise reduction method and device, computer equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410676009.8A CN118296307A (en) | 2024-05-29 | 2024-05-29 | Logging data noise reduction method and device, computer equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118296307A true CN118296307A (en) | 2024-07-05 |
Family
ID=91686570
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410676009.8A Pending CN118296307A (en) | 2024-05-29 | 2024-05-29 | Logging data noise reduction method and device, computer equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118296307A (en) |
-
2024
- 2024-05-29 CN CN202410676009.8A patent/CN118296307A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhu et al. | Intelligent logging lithological interpretation with convolution neural networks | |
| CN110212528B (en) | Power distribution network measurement data missing reconstruction method | |
| CN111562612B (en) | Deep learning microseismic event identification method and system based on attention mechanism | |
| CN110807544B (en) | Oil field residual oil saturation distribution prediction method based on machine learning | |
| CN112926771B (en) | Oil yield prediction method and system based on improved space-time diagram convolution network | |
| CN110726898A (en) | Power distribution network fault type identification method | |
| CN115146700A (en) | Runoff prediction method based on Transformer sequence-to-sequence model | |
| CN117556369B (en) | Power theft detection method and system for dynamically generated residual error graph convolution neural network | |
| CN114595427B (en) | Vehicle track sequence filling and repairing method containing missing values based on non-autoregressive model | |
| CN116485029A (en) | Method, device, equipment and medium for detecting accuracy of dynamic data in oilfield development | |
| CN114860709A (en) | Bi-GAN-based power system missing value filling method | |
| CN114818817A (en) | Weak fault recognition system and method for capacitive voltage transformer | |
| CN113536508A (en) | Method and system for classifying manufacturing network nodes | |
| CN117349657A (en) | Distributed data acquisition module and monitoring system for hydraulic engineering environment monitoring | |
| CN117079005A (en) | Optical cable fault monitoring method, system, device and readable storage medium | |
| CN118296307A (en) | Logging data noise reduction method and device, computer equipment and storage medium | |
| CN114299330A (en) | Seismic facies classification method | |
| CN115374687A (en) | Numerical-shape combined intelligent diagnosis method for working conditions of oil well | |
| CN112346126A (en) | Low-order fault identification method, device, equipment and readable storage medium | |
| CN117909691B (en) | Ocean engineering design data acquisition system and method | |
| CN117975174B (en) | Three-dimensional digital core reconstruction method based on improvement VQGAN | |
| CN117574327B (en) | Fracturing pump fault detection method, system and storage medium | |
| CN112098291A (en) | Electric imaging porosity spectrum-based permeability calculation method and system | |
| CN118033732B (en) | Seismic data reconstruction method based on airspace frequency domain fusion architecture | |
| CN113554106B (en) | Collaborative completion method for power missing data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination |